trunk/2ed93d5582b2e8d1d8a401de37e38f2b3e17649f: Add operator microbenchmark comparison workflow for PRs (#179476)

Add a new GitHub Action (operator_microbenchmark_compare.yml) that benchmarks a single operator on a chosen GPU from a PR branch, compares results against the latest daily CI run on main, and posts a formatted comparison table as a PR comment.

The workflow reuses the existing _linux-build.yml and _linux-test.yml reusable workflows, following the pattern established by inductor-perf-compare.yml. It supports H100, A100, B200 (CUDA) and MI300X (ROCm) GPUs.

To enable single-operator selection, test.sh now parses the test config name (e.g. operator_microbenchmark_matmul_test) to extract the operator, and the test_operator_microbenchmark function respects the OP_BENCHMARK_TESTS env var. This is fully backward-compatible: the existing operator_microbenchmark.yml workflow uses config "operator_microbenchmark_test" which falls through to the default full operator list.

Pull Request resolved:

Co-authored-by: Huy Do [email protected]