Updating version for rubygems for latest - #1171

Fixes consistent, reproducible failures observed across 2 consecutive
CI runs on this PR, all with the same signature:

  Timed out after 20.001s.
  Get "http://localhost:XXXXX": dial tcp [::1]:XXXXX: connect: connection refused

Traced this to switchblade's Docker platform backend
(internal/docker/start.go): ContainerStart()+ContainerInspect() return
the published port as soon as the container itself is reported
started, with no wait for the app process inside to actually finish
booting and bind that port. The default 20s Eventually timeout
(already more generous than sibling buildpacks' 10s default, since
Ruby/Rails apps boot slower) isn't always enough headroom when many
containers start in parallel on a resource-constrained CI runner
(these all failed specifically in the "parallel=true" matrix legs).

Applied targeted per-test overrides (90s timeout, 2s poll interval)
to exactly the 3 consistently-failing tests, rather than changing the
global default - this matches the established precedent in this repo
for this exact class of issue (#1140, which did the same for the
JRuby fixture) rather than slowing down failure detection for every
other test in the suite.

Affected tests:
- Offline/runs_a_vendored_cached_app
- Default/vendor_bundle
- Default/custom_gemfile