Point parallel_rspec at the cached spec runtime log

[gavinomelia]

Summary

parallel_rspec reads tmp/parallel_runtime_rspec.log by default, but rails-ci.yml and both composite actions restore the cached runtime log to tmp/turbo_rspec_runtime.log — the path rolemodel_rails' generator writes into every app's .rspec_parallel. It therefore never found the runtimes and silently fell back to grouping specs by file size, so worker processes finish minutes apart while cores sit idle.

Changes

• Pass --runtime-log so parallel_rspec actually reads the cached runtimes
• Apply the same fix to both legacy composite actions
• Rename the two cache steps, labelled "Turbo-Test" but feeding parallel_rspec
• Add a CHANGELOG entry for v3.11.3

Measured on CI

Verified end to end on ServeWell (blacksmith-16vcpu, 16 processes) by pinning its ci.yml at this branch. Baseline is the five most recent successful main/feature runs; the fix is run 33420158397.

	baseline (5 runs)	with fix
rspec step	353s mean (336–371)	267s	−86s, −24%
slowest worker	349s mean	264s	−85s
worker spread	~208s	46s	−78%
idle core-seconds	29%	8%
total work	3959s	3902s	unchanged

Total work is flat, so this is redistribution rather than the suite getting cheaper. The fixed run prints Using recorded test runtime; none of the five baseline runs do.

The residual 8% is a floor from indivisible files — ServeWell has a single 191s example (spec/jobs/generate_packet_job_spec.rb) so no worker can finish before ~218s.

Notes

• Safe with no cache: parallel_tests rescues the missing file and falls back to file size, exactly as today.
• All 34 RoleModel repos with a .rspec_parallel write to tmp/turbo_rspec_runtime.log, so the path is safe to hardcode.
• Gains will differ per app — they depend on how uneven the suite is and how badly file size correlates with runtime.
• Complementary to Fix turbo-test runtime cache never being saved #18, not a duplicate — that PR adds the missing cache save to the composite actions. Neither alone fixes them: Fix turbo-test runtime cache never being saved #18 without this flag still groups by file size, and this flag without Fix turbo-test runtime cache never being saved #18 has nothing to restore. They merge without conflict. (Fix turbo-test runtime cache never being saved #18 is also 11 commits behind main and still pins actions/cache@v5.)