test: document container inference bugs with Any bounds

[asukaminato0721]

Summary

Any-bounded type variables erase concrete container element types and over-narrow repeated appends. The tests retain the intended type assertions and explicitly expect the current incorrect diagnostics.

The testcase! regressions use bug = markers and comments documenting the correct behavior. These are passing tests of known bugs, following the repository test guidance.

The complete implementation and regression tests remain in PR #4877. Related issue: #3265.

Validation

Based on upstream main at ef6de648bce0dc6dcca0a21a9fbb7490f23ecce8.

• cargo test -p pyrefly --lib -- test_iter_container_literal test_any_bounded_typevar_container_hints test_any_bounded_typevar_empty_list_repeated_append — passed.
• python3 test.py --no-test --no-tensor-shapes --no-conformance --no-jsonschema --no-extension — passed.

[meta-codesync]

This pull request has been imported. If you are a Meta employee, you can view this in D122390321. (Because this pull request was imported automatically, there will not be any future comments.)

[rchen152]

It looks like some of the tests you updated are failing; could you take a look?

[rchen152]

The tests assert the intended behavior and deliberately fail on current main until the fix is applied.

Oh, I see - sorry for not reading properly!

The way to write deliberately failing tests is to make the test pass on main but add a bug= marker (https://github.com/facebook/pyrefly/blob/main/AGENTS.md#the-bug-marker-in-tests) and comments documenting that the test demonstrates a bug and what the correct behavior would be.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅