test: Align QA BF16 with ml_dtypes and generate ONNX BF16 models

[yinggeh]

What does the PR do?

Aligns server QA with the Python client change for BF16: BF16 I/O uses ml_dtypes.bfloat16 instead of float32 truncation.

Extends the QA model factory so ONNX BF16 models are generated.

L0_backend_onnxruntime now copies the generated onnx_bf16_bf16_bf16 model from the QA data directory and exercises add/sub with batched BF16 inputs; removes the standalone gen_add_bf16_onnx_model.py generator.

Checklist

• PR title reflects the change and is of format <commit_type>: <Title>
• Changes are described in the pull request.
• Related issues are referenced.
• Populated github labels field
• Added test plan and verified test passes.
• Verified that the PR passes existing CI.
• Verified copyright is correct on all changed files.
• Added succinct git squash message before merging ref.
• All template sections are filled out.
• Optional: Additional screenshots for behavior/output changes with before/after.

Commit Type:

Check the conventional commit type
box here and add the label to the github PR.

• test

Related PRs:

triton-inference-server/client#897

Where should the reviewer start?

Test plan:

• CI Pipeline ID:
  51358164

Caveats:

Background

Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)