Feature/genesis extension

[Ahmedhasssan-aig]

Summary

Adds a Genesis/Taichi physics-simulation profiling extension to TraceLens. It combines rocprof kernel analysis with pftrace HIP-activity and memory-copy reports, categorizes physics kernels, and isolates the steady-state simulation window (excluding JIT/build overhead) for accurate performance numbers.

What's new

• New report generator TraceLens_generate_perf_report_genesis that consumes a rocprofv3 capture directory and emits a physics-focused performance report.
• Physics kernel categorization — groups kernels by role (Rigid Body Solver, Collision, Time Integration, etc.).
• Steady-state isolation — automatically splits JIT/build overhead from the actual simulation window so metrics reflect real runtime.
• Multi-source analysis — merges kernel dispatches (CSV), HIP activity + memory copies (pftrace), with a native rocprof JSON fallback.

Inputs

Captured via rocprofv3 --hip-trace --kernel-trace --memory-copy-trace --output-format pftrace. Expects profile_output//kernel_trace/ (requires kernel_kernel_trace.csv; uses kernel_results.pftrace/kernel_results.json as available), with optional run.log (auto-detects benchmark wall-time) and combined_manifest.json (capture metadata).

Outputs

• genesis_perf_report.xlsx — GPU timeline, kernel summary by physics category, HIP activity, and memory-copy sheets.
• genesis_summary.md — steady-state GPU utilization overview + top kernels.

Key options

• --steady-state-gap-ms (default 1000) — gap threshold to split JIT from simulation.
• --steady-state-fallback-s — override the timed window (default auto-detected from run.log).
• --include-api — include HIP/HSA API events in the rocprof JSON.
• --keep-work — retain intermediate files for debugging.

[gabeweisz]

Looks pretty good and will not break anything, but we need to add some tests and fix the failing doc creation