Benchmark, package, and verify v0.5.0

[itdevwu]

Goal

Release CPU/Python inventory only after its usefulness, bounds, perturbation, compatibility, and installation paths are measured.

Acceptance criteria

• Benchmarks compare broad CPU inventory with perf and py-spy separately, then compare narrowed exact xprobe collection against the broad phase.
• Results report workload throughput, collector CPU/RSS, target RSS, sample loss, stack coverage, symbol coverage, capacity, and artifact size without claiming universal superiority.
• Native, Python 3.12+ perf-map, syscall aggregate, CPython GC, mixed CPU/GPU, BPF cleanup, and agent-contract tests pass as applicable.
• The hardware workflow runs on a registered runner or an explicitly selected maintained provider and is a visible release gate.
• GitHub Actions Node runtime deprecation warnings are removed by supported action versions.
• Linux x86_64 release artifacts retain the GLIBC_2.34 ceiling and CUDA 12/13 Agents, and the public archive passes checksum, installation, and ELF inspection.