[None][perf] Widen MiniMax-M3 decode q in-register instead of in a copy

[brb-nv]

The fused QK-norm+RoPE producer emits E4M3 q, but the ported Triton sparse decode kernel derives its compute dtype from q and dequantizes the paged cache into it, so the caller materialized a widened BF16 copy first. That copy is a standalone kernel between the block selector and the decode kernel, where nothing overlaps it: ~3.5 us on every sparse layer, ~0.2 ms per decode step at TP4 on B300.

Widen inside the kernel instead, immediately after the load and before anything reads q.dtype. Doing it later would leave K, V and the softmax probabilities cast to E4M3 and run the whole attention in FP8. E4M3 -> BF16 is exact, so the result is bitwise identical to the copy it replaces, which the new test asserts as an equality.

The M3 accuracy test now pins indexer_kv_dtype=fp8 and fuse_qkv_index_projection on the MSA path, which is what makes the producer emit E4M3 q at all; without it this widen never fires under test.

@coderabbitai summary

Description

Test Coverage

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (see test instructions)

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• Update tava architecture diagram if there is a significant design change in PR.

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.