Add vein estimator validation tools and benchmark record

[Toti330]

Summary

• add a database-free exact vein exporter and deterministic candidate fixtures
• add locked corpus validation, held-out protection, accuracy gates, scale fitting, and elapsed-overhead reporting
• publish raw calibration and baseline artifacts plus a concise Markdown benchmark timeline

Scope

• no production generation behavior changes
• the recorded spacing estimator is rejected: it misses P95 and per-ore WAPE requirements
• held-out data remains unopened and candidate performance remains unmeasured because calibration failed

Validation

• PYTHONPATH=benchmarks python3 -m unittest benchmarks/test_evaluate_vein_accuracy.py
• cargo test --release --locked in the pinned Rust 1.97.0 builder
• JSON parsing and git diff --check
• held-out exporter guard rejects unapproved access

Formatting Note

rustfmt --check reports existing formatting drift in shared algorithm modules and an existing section of generate_csv.rs; this PR does not reformat unrelated production code.

[Toti330]

The original goal was to increase seed-generation throughput without changing deterministic output, a current run has produced reproducible validation evidence, and that is a point where an unsuccessful approximation should be documented rather than promoted. The agent was not able to produce a vein estimator that met the predeclared accuracy targets, so production generation remains unchanged and the exact same as before.