P0-1: Standard getpid benchmark (baseline/uprobe/lightprobe-entry/ret)
  - WSL2: entry overhead 293ns (<1000ns), uprobe N/A
  - VM: entry overhead 1259ns, uprobe 2898ns (lightprobe 57% faster)
  - target_getpid_bench.c with --label/--wait/JSON output
  - run_perf_benchmark.sh with tracefs uprobe fallback

P0-2: attach/detach latency (~38ms CLI end-to-end)
  - run_latency_bench.sh for getpid/malloc/strlen

P0-3/F6: 16 concurrent probes on dynamic library
  - libmulti_probe_target.so (f01-f16)
  - run_16probe_smoke.sh: 16/16 attach/events/detach/clean

P1: Signal stress test
  - target_signal_stress.c: 4 threads + high-freq SIGUSR1
  - run_signal_stress.sh: entry=return=2048, no crash

P2: XMM0-XMM15 register save/restore
  - xmm_save_restore_x86_64.S + stub modifications
  - fp_probe via shared library (libfp_probe_target.so)
  - Delta < 1e-12 verified

Docs & demo:
  - demo.sh: one-click 6-step validation (10/10 checks)
  - benchmark_report.md: dual-environment data + uprobe comparison
  - improvement_plan.md: execution status tracking
  - README.md: updated quickstart, password instructions
  - LICENSE: Apache 2.0

Engineering:
  - State file: reverted to global /tmp/lightprobe_state.bin
  - context_x86_64.S: added clarification comment
  - All scripts use LIGHTPROBE_SUDO_PASSWORD env var
  - Removed VM-specific scripts with hardcoded credentials