Add reproducible benchmark JSON artifacts

[helly25]

Add one strict, reusable measurement envelope for every Arena, SegmentedSequence, HAMT, and StringInterner benchmark. This makes raw performance evidence attributable, comparable, and ready for later CI and artifact-store ingestion.

AG;DR

• Adds a standard-library-only runner that executes Google Benchmark with nine repetitions, random interleaving, numeric-seconds warmup, and minimum-time controls.
• Records raw benchmark rows plus full Git, host, CPU, compiler, Bazel, C++ baseline, command, load, and timing provenance.
• Resolves Apple Silicon chip identity through the hardware profiler before accepting generic arm processor text, without storing serial-number or UUID fields.
• Adds explicit valid/suspect/invalid classification so surprising runs are retained rather than silently discarded.
• Adds a content hash and strict validation that rejects weak controls, empty results, incomplete provenance, and modified artifacts.
• Publishes the versioned JSON Schema for later Artifactory ingestion and chart generation.
• Integrates focused regression tests into pre-commit, including exact numeric warmup formatting and Apple chip detection.

Stack:

• Depends on Define production string interning architecture #438.
• Must not merge before the implementation stack has reference-machine measurement evidence.

Validation at 8a2d2f155:

• python3 -m unittest tools/benchmark_artifact_test.py (7 tests)
• pre-commit run --files tools/benchmark_artifact.py tools/benchmark_artifact_test.py tools/benchmark_artifact.schema.json .pre-commit-config.yaml
• git diff --check