Generate perf reports from one YAML config

[ajassani]

Summary

• Add python -m TraceLens.Reporting.generate_perf_report_from_config --config run.yaml, also installed as TraceLens_generate_perf_report_from_config.
• The YAML is validated before any trace is opened. default calls the training report. inference_graph_capture, spec_dec, and pd_disaggregation call the inference report. Shared report kwargs are top-level and apply to every job. platform is a bundled arch name resolved through load_arch.
• Unless the YAML turns them off, enable_pseudo_ops, group_by_num_kernels, include_call_stack, and group_by_parent_module default on when the selected function accepts them.
• spec_dec and pd_disaggregation still call the inference report. spec_decode (method, num_spec_tokens) is checked and kept on the plan. PD writes perf_report_<role>_rank<r>. Comparison stays a separate step: one wrapper call per trace, then compare.

Test plan

• pytest tests/test_perf_report_from_config.py (28 tests: schema, dispatch, recipe defaults, PD filenames, capture-merge hook, and a failed second job never calls the report function)
• Run one training trace and one inference capture trace through a YAML and compare the sheets to the existing CLIs

[codecov-commenter]

Codecov Report

❌ Patch coverage is 90.47619% with 20 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...Lens/Reporting/generate_perf_report_from_config.py	90.47%	20 Missing ⚠️

📢 Thoughts on this report? Let us know!