feat(replan): deliver core Goal and dense typed decision evidence

[huangruiteng]

Replan could require an agent to discover a separate evidence command and guess its JSON fields, while the delivered context kept only a small recent coverage window. The current obligation now carries the core Goal, scoped acceptance, and dense decision evidence through the existing TypeScript work-item owner.

• Retire evidence-log without an alias; keep historical receipt decoding and internal supervisor/recovery readers. Exact optional reads use bound history --goal-id ... --agent-id ... --evidence-ref ... actions and fail on stale or wrong-scope references.
• Reduce repeated observations before selecting up to 24 distinct results/routes from the complete compact index. Show omitted coverage explicitly; novelty admission still checks complete available history. Read receipts and unchanged/prose-only claims gain no settlement authority.
• Preserve stable obligation assignment and receipt-bound Agent identity through settlement. Full review packets carry structured context; handoff forwarding avoids duplicating it.

This intentionally changes the CLI and replan projection defaults. It adds no provider, setting, frontend screen, Lark operation, or write authority. The existing quota/review-packet journey owns delivery; the Python codec only adapts existing Goal/history sources and the private snapshot transport. Durable legacy receipts remain readable, with faithful old-format negative fixtures rather than invented command placeholders.

Validation: focused Python run 321 passed, 2 failed; both failures in test_replan_successor_durable_ack.py (file/SQLite periodic successor closeout) reproduce on untouched b21929c4f16c13530aa270a4bfbf082ed8d925dc. Subsequent affected receipt/blocker tests, real CLI reference/retirement checks, quota/review/handoff smokes, 44 TypeScript tests, control-plane typecheck, Ruff, semantic vocabulary and registry-I/O checks passed. Base/head CLI differential passed with the intentional migration signals. Rebased onto e55489c7791f326abe63e7b9b2952a134d652cf7; the incoming Lark change has no overlapping files.

The same crowded required-replan fixture grows 26,844 → 33,523 characters; ordinary quota fixtures are unchanged and handoff-only output shrinks. The documented budget increase is restricted to decision context, not a general guard allowance. These measurements do not prove token savings or long-horizon score gains. This is a reviewable S3/S6 increment; controlled long-horizon quality/cost qualification remains S11 work. No benchmark or live-model jobs were launched.

The production frontend bundle built and passed source-provenance verification. The isolated real installer smoke passed installed CLI help, generated/compressed manpage readback and command-catalog retirement. Post-integration registry-I/O, semantic checks and 16 real CLI/reentry cases passed. Final CI is pending; this runtime/control-plane PR is left for maintainer review and merge.

[mergify]

This pull request has merge conflicts with main and cannot be merged
until they are resolved. Please rebase or merge the base branch, @huangruiteng.

Choose the remote for the base repository, not an out-of-date fork.
For a fork clone, first inspect git remote -v; upstream must point
to https://github.com/loopx-project/loopx.git. If it is absent, add it
with git remote add upstream https://github.com/loopx-project/loopx.git.
Then run:

git fetch upstream
git rebase upstream/main
# Resolve each conflict, git add the resolved files, then git rebase --continue.
git push --force-with-lease origin HEAD

For a same-repository clone whose origin points to
https://github.com/loopx-project/loopx.git, use origin instead of
upstream for fetch/rebase. If you prefer merging the base, use
git merge <base-remote>/main and push normally.

Keep the DCO Signed-off-by trailer on every commit when you rebase.
https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork