Fix match_bits on bfloat and drop strip_widening_cast integer guard

[alexreinking]

match_bits widened the narrower operand with Type::with_bits, which on a bfloat16 produced a bfloat32 - a type codegen can't handle. Reachable directly, e.g. widening_mul(bf16, f32). Widen a bfloat to a regular float instead (matching the wider operand's bit count), consistent with Type::widen().

With that fixed, the is_int_or_uint guard in strip_widening_cast is no longer needed to avoid synthesizing a bfloat32, so drop it. bfloat16 inner types are now stripped like other floats, so f32(bf16) * f32(bf16) is recognized as a widening_mul, matching the existing float16 handling.

Add a deterministic codegen regression test for widening_mul(bf16, f32), the widening_mul(bf16, bf16) case, and give the lossless_cast fuzzer an isolated float/bfloat16/float16 expression pool (kept separate from the integer pool, since float->int casts of unbounded floats are UB). The fuzzer is what surfaced the match_bits bug.

Checklist

• Tests added or updated (not required for docs, CI config, or typo fixes)
• Commits include AI attribution where applicable (see Code of Conduct)

[abadams]

Failures appear to be real

[alexreinking]

Failures appear to be real

Yeah, seems this broke some pattern the X86 backend depended on. Looking into it.

[abadams]

The changes to the lossless cast fuzz test don't make sense to me - that test compares the output for exact equality after casting to int64, which is simultaneously too precise and not precise enough for floating point transformations. I think the new bfloat pattern should just be in simd_op_check.

[alexreinking]

The changes to the lossless cast fuzz test don't make sense to me - that test compares the output for exact equality after casting to int64, which is simultaneously too precise and not precise enough for floating point transformations.

It doesn't check equality for floating point though... just that it doesn't assert-fail the compiler. The previous widening logic produced a BFloat(32) that was nonsensical and crashed.

I think the new bfloat pattern should just be in simd_op_check.

It's already in the diff.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@5476bf3). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #9218   +/-   ##
=======================================
  Coverage        ?   69.80%           
=======================================
  Files           ?      255           
  Lines           ?    78580           
  Branches        ?    18790           
=======================================
  Hits            ?    54849           
  Misses          ?    18167           
  Partials        ?     5564           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.