Clean up CI rot: sorbet and vendored RBI noise

[justin-reid]

Summary

Clear the red typecheck job that has been pre-existing on main against current Sorbet. Rebased on top of #41 (now merged); the Lint/UselessAssignment fix that originally lived here landed independently in #41 (78e37f6) and was dropped on rebase, so this PR is sorbet/RBI-only.

After this PR, bin/srb tc reports 0 errors and bin/rubocop reports no offences. Existing tests unchanged (69 runs / 337 assertions, all pass on dev test).

Changes

1. sorbet/typed_overrides.yaml (new) — downgrades 17 autogenerated sorbet/rbi/gems/*.rbi files that ship at # typed: strict / true but omit sigs on many methods (~4360 sig-less-method errors of pure vendored noise). Editing the sigil in place would be wiped on the next srb rbi gems regen; the yaml + --typed-override in sorbet/config survives regeneration.

2. sorbet/config — adds --typed-override sorbet/typed_overrides.yaml.

3. sorbet/rbi/shims/prism.rbi (new) — defines the Prism::LexCompat::Result namespace so a stale sig in prism-1.9.0's gem-shipped RBI (returns(Prism::LexCompat::Result)) resolves. LexCompat was renamed to LexResult upstream but the sig wasn't updated. Per @jose-shopify's review on r3350789249, this replaces an earlier --suppress-error-code=5002 in sorbet/config — --suppress-error-code is global, so the shim is the narrower fix that keeps 5002 firing for our own lib//test/ code.

4. sorbet/rbi/minitest.rbi — demotes # typed: strict → true (it omits sigs on ~17 hand-written methods), and fixes two existing sigs that this repo's own tests exercise:

   • assert_includes's collection is T.untyped, not T::Enumerable[T.untyped] — Minitest dispatches via .include? and accepts String (used by redis_coordinator_integration_test.rb:362-364).
   • assert_raises takes a block; the sig was missing &blk (used by defined_runnable_test.rb:13).

5. lib/minitest/distributed/coordinators/redis_coordinator.rb — T::Array[Module] on the custom_middlewares accessor → T::Array[T.untyped]. Sorbet 5046 requires Module to carry type arguments; T::Module[…] would type-check but raises NameError: uninitialized constant T::Module at load time on sorbet-runtime 0.5.12443 (the version CI resolves), so T.untyped is the runtime-safe choice.

How to test

Manual validation:

• dev up (boots redis service)
• shadowenv exec -- bundle exec rake test → 69 runs / 337 assertions / 0 failures
• shadowenv exec -- bin/srb tc → No errors! Great job.
• shadowenv exec -- bin/rubocop → 58 files inspected, no offenses detected

CI: lint, typecheck, and tests (2.7/3/3.1/3.2) all green on the rebased branch.

[jose-shopify]

left a comment about suppressing the 5002 err

[jose-shopify]

Heads up: The flag isn't scoped to prism, it's global. It doesn't say "ignore this in prism's file," it says "ignore 5002 everywhere, forever." So if someone later writes a typo'd class name in this project's own lib/ or test/ code, Sorbet would normally catch it, but now it stays silent.

[justin-reid]

Good catch — fixed in 2d98516.

--suppress-error-code=5002 is gone from sorbet/config. Replaced it with a 13-line shim at sorbet/rbi/shims/prism.rbi that defines the missing Prism::LexCompat::Result namespace so prism-1.9.0's stale sig (the only 5002 source in the repo, verified with bin/srb tc --isolate-error-code 5002) resolves. 5002 is now back to firing for our own lib//test/ code — a typo'd class name will be caught again. Typecheck stays clean: No errors! Great job.

_{🤖 AE · ⚠️ unapproved by @justin-reid}