[sunzenshen]

Description

Bots climb down ladders instead of bouncing off the top

Depends on #2176 ("Stop bots fighting the engine over unwanted ladders") and is stacked on its branch; review the
last three commits. With #2176 the climb behavior claims the ladder for PlayerLocomotion, whose descend states then
steer every descent the behavior starts. This PR makes those states, and the two ladder behaviors around them, take a
bot down.

A bot that starts down a ladder from its top usually ends up back on the floor it started from, and tries again.
ntre_saitama_ctg's elevator hatch shows it most: 95 % of the descents of the ladder in the hatch end back on the car
roof, about 25 attempts a match, with or without #2176 (94 % on upstream's ladder code). The bots then walk a
floor-level path on the roof into its walls, the map's largest stuck spot. Three things put them back on top:

1. Forward climbs up. CGameMovement::LadderMove() turns the part of a move that pushes into the ladder's face
   into climbing up, and moves the player along the face with the rest. Facing the ladder, forward goes down only
   while the view points more than 45 degrees down; with the view level it climbs. Every descent pressed forward before
   the view was down: the approach's last forward press stays held until the next bot update, after the grab;
   ApproachDescendingLadder() kept steering towards its mount point after the bot was already on the ladder; and
   Approach() presses forward on any ladder the bot is descending. DescendLadder() aimed away from the face while
   the climb behavior aimed into it, both at MANDATORY priority, and PlayerBody refuses a same-priority aim until the
   head has been steady for 0.3 s, so the view was slow to come round at all. Near the top, a few ticks of climbing
   lift the bot past the ladder's top and off it.
2. Grabbed while still standing at the top. The engine puts a bot on the ladder as soon as its move points at it,
   on the landing or on top of the ladder brush, which is solid and often stands a step proud of the floor (the hatch
   ladder's top is 14 u over the roof). From there no press takes the bot down: into the face pushes into what it
   stands on, and a move away from the face on a floor makes LadderMove() push it off the ladder.
3. The early jump-off lands it back on top. CNEOBotLadderClimb jumps off as soon as the exit area is within a
   safe drop below, which on a short ladder is at the mount, and kicks the bot towards the exit area's center. Where
   the top floor overhangs the exit (the roof over the car), that lands it back on the top floor.

Changes, in three commits

1. Press forward down a ladder only when it descends (NextBotPlayerLocomotion.{h,cpp}, neo_bot_ladder_climb.cpp).
   PlayerLocomotion::IsForwardDownLadder() says whether a forward press moves the bot down, from the view and the face
   it holds (the ladder brush's back face when it grabbed the ladder from behind); a press facing away from the face does
   not count. Approach() presses nothing on a descent until it does, and DescendLadder() looks down into the face
   rather than away from it. ApproachDescendingLadder() passes a bot already on the ladder straight to
   DescendLadder(), before its "within a jump of the bottom" check lets go of it at the top of a short ladder. The
   climb behavior gates its own forward press the same way, does not count the wait for the view as a stall, and leaves
   the aim of a claimed descent to the locomotion.
2. Mount a descent from in front of the ladder (neo_bot_ladder_approach.{h,cpp}). The approach heads for where a
   descending bot hangs, half a hull plus 2 u in front of the face at the ladder's top, looks down the ladder rather
   than level at its middle, and mounts only once the bot is off the top (below the ladder's top, and no longer standing
   within a step of it). It lines up on the ladder's normal line only from in front of the face, where coming in at an
   angle leaves the bot on the floor beside the drop, stepping up onto the ladder's top; from beside or behind the
   ladder it walks straight on, over the top.
3. Keep a descent's early jump-off below the top (neo_bot_ladder_climb.cpp). Going down, the jump-off waits until
   the bot is a standing height below the ladder's top.

No convar switches any of this off. Climbing up is unchanged.

Evidence

Upstream at #2168 plus #2176 and the draft SeekWeapon PR (#2186), with harness logging added and a test-only switch
for this change; both arms run the same server build. 30 bots, default weapon damage, 8 matches at a time; each run is
paired with a run of the other arm on the same map, ghost spawn and repeat, back to back. CTG: 2 rounds of 3 minutes with
the ghost pinned at one ghost spawn; TDM and JGR: one 5-minute round. All 384 matches below reached match end with 0
asserts and no crash. "Stuck" is OnStuck plus FAIL_STUCK move failures, without the round-start spawn crowd and
weapon scavenging; intervals are 95 % bootstrap intervals of the paired ratio. A descent is a climb down the behavior
starts; it "ends back on top" when the bot is within 24 u of the ladder's top when the climb ends.

ntre_saitama_ctg, 68 pairs: every ghost spawn twice, on master's navmesh and on the reworked one from the
companion neoAssets PR (NeotokyoRebuild/neoAssets#146):

a match	#2176	+ this PR	paired
hatch ladder descents	25.1	2.9
... ending back on the roof	95 % (1630 of 1706)	5 % (10 of 200)
stuck	41.2	30.2	x0.71 [0.60, 0.87], 44 down / 23 up
stuck in the ladder tower	19.2	6.5	x0.35 [0.24, 0.53]
stuck elsewhere	22.0	23.7	x1.11 [0.93, 1.31]
FAIL_NO_PATH_EXISTS	4.5	4.2
kills	49.3	49.3	x1.00 [0.98, 1.02]
fall and hurt-volume deaths	0	0

• Bots that reach the bottom stop retrying, so the hatch ladder sees a tenth of the descents it did.
• The map's other short ladders: 32 descents, 18 back on top (56 %) with More tweaks to bot ladder approaches, climbs and dismounts #2176; 6 descents, all to the bottom, with
  this PR.
• "Stuck elsewhere" is x1.11 on both navmeshes (each interval across 1): the bots the tower no longer holds reach the
  map's other spots, on master's navmesh mostly its 3492 / 3493. Capped at 20 stuck events a bot, the total is x0.74
  [0.62, 0.88].
• Against the same code without the SeekWeapon PR, the same pairs measure x0.79 [0.64, 0.99] (tower x0.39 [0.26,
  0.57]): the bots that PR keeps out of the fire pit no longer end up in this trap, which its own measurement showed as
  a rise in other stuck (x1.27 [0.98, 1.67]).

Other ladder maps, 72 pairs: 5 ghost spawns twice on transit, tarmac, subsurface, marketa, rogue and ghost CTG,
yard JGR's 3 juggernaut spawns twice, 6 bullet TDM matches (the navmeshes of their companion or planned neoAssets PRs;
ghost and yard on their current research navmeshes):

	#2176	+ this PR
descents a match	1.9	0.4
... ending back on top	82 % (112 of 137)	0 % (0 of 32)
climbs up a match	11.8	12.0
stuck, paired		x0.92 [0.72, 1.16], 35 down / 35 up
"fell off the path"	3.2	2.6 (x0.81 [0.67, 0.98])
FAIL_NO_PATH_EXISTS	6.5	6.4
kills	67.1	66.7 (x0.99 [0.94, 1.03])
fall and hurt-volume deaths	9	10 (rogue's canal, bullet's track)

By map (10 pairs; 6 for yard and bullet), every interval across 1: yard x0.57, bullet x0.72, subsurface x0.77, rogue
x0.82, marketa x0.83, tarmac x1.10, transit x1.25, ghost x1.27. Ghost has no ladder descents; its figure is one
match in which three bots logged 68-138 stuck events each (one of those spots piles up the same way in the #2176
arm). On transit the descent of ladder 2 now works (8.7 attempts a match, 94 % back on top, -> 2.6, 0 %), and more
bots reach the pilot car below it, a known stuck spot (625: 27 -> 63 events in 10 matches).

Descents from behind the ladder's plane. ntre_bullet_tdm's shaft ladders have forward tops, landings behind the
ladder's plane, which its reworked navmesh keeps as climb exits only, because descents from there failed.
On master's navmesh (8 matches an arm, one round each): ladder 1 19 of 20 descents back on top with #2176, 3 of 5
with this PR; ladder 4 27 of 27, then 0 of 8.

Cost

Two dot products per bot update while a bot descends a ladder.

Risk

Moderate. Every ladder descent changes: the approach now takes the bot to the point it hangs from and mounts only once
it is off the top, and the locomotion holds still until the view points down. Where the walk out to that point is
blocked, the approach's existing 3 s timeout ends it, as before.

Toolchain

• Linux GCC Distro Native (Linux Mint 22.3, GCC 13.3.0)

Linked Issues

• related More tweaks to bot ladder approaches, climbs and dismounts #2176
• related Bots follow their path on NAV_MESH_PRECISE areas #2168
• related Bots skip weapons they cannot reach when scavenging #2186
• related ntre_saitama_ctg: navmesh rework neoAssets#146

[sunzenshen]

Closing this in favor of combining with #2176, as these both cover the same ladder testing case and should be reviewed together.