[codex] stabilize trelu golden dir mapping

[zhoubot]

What changed

• replace the index-based i < 8 directory selection in trelu/gen_data.py with explicit per-case output names
• map the BF16 case through the same explicit mechanism instead of relying on loop position

Why

The previous logic depended on the current ordering and size of case_params_list. Any insertion, deletion, or reordering could silently break the mapping between generated golden-data directories and the gtest names used by TRELUTest.

Impact

• trelu golden-data generation is now tied to the intended gtest case names instead of fragile list indexes
• future testcase additions or reordering will not accidentally break the directory layout expected by the CPU-SIM test binary

Validation

• python3 -m py_compile tests/cpu/st/testcase/trelu/gen_data.py
• ruff check tests/cpu/st/testcase/trelu/gen_data.py
• python3 tests/run_cpu.py --testcase trelu --gtest_filter 'TRELUTest.case_0' --generator Ninja --build-dir build/cpu_pr_trelu_dir_mapping --clean

[gemini-code-assist]

Code Review

This pull request refactors the TRelu test data generation script by introducing an explicit output_case_name attribute to the TReluParams class, which is now used for directory naming instead of loop indices. A review comment suggests that the case_name variable and the generate_case_name function have become redundant following these changes and should be removed to simplify the script.

[gemini-code-assist]

The case_name variable (defined on line 81) is no longer used for directory mapping, as it has been replaced by the explicit param.output_case_name. Additionally, case_name is passed to gen_golden_data_trelu (line 86), which does not use the parameter. This makes the generate_case_name function and the case_name variable redundant. Consider removing them to simplify the script and improve maintainability.