Refactor benchmarking to self-contained C implementation

- Add mknes_bench.c with direct PMC access via perf_event_open()
- Remove dependency on external perf/awk for statistics
- Add RT priority and CPU affinity control in C code
- Use static BSS allocation (page-aligned) instead of malloc
- Add stalled-cycles-backend counter (gracefully handles AMD unavailability)
- Add throughput metrics (MIPS, Mcycles/sec)
- Optimize Bench.sh to only regenerate profile data when needed
- Add -n and -f flags for configurable runs and frames
- Suppress mapper messages during benchmark
- ~6x faster benchmark workflow (20s first run, 16s subsequent)

0 files changed, 0 insertions, 0 deletions