[GGGP M4] Release readiness — CI, semver, benches, docs, v0.1.0

[mnemonik-dev]

Summary

Milestone M4: make evoforge-grammar production-ready / publishable (mirrors evoforge's
Implementation-Plan Milestone 9). Lock the public API + formats, add CI, benchmarks, audit, docs, and
tag v0.1.0. This is the last pure-Rust milestone; the Python bridge + real Archimedes fitness are a
separate later phase.

Spec: https://github.com/mnemonik-dev/genetic_algorithms/blob/main/docs/technical-reference/gggp-implementation-spec.md (§7 release gates, §8 M4)

Scope

• CI (.github/workflows): cargo fmt --check, cargo clippy --all-targets --all-features -D warnings,
  cargo test --all-features, cargo doc --no-deps, feature-matrix build.
• Semver/stability doc: declare stability of the grammar JSON format and the
  snapshot() format (the deterministic-snapshot contract); document MSRV (Rust 1.75).
• Benchmarks: decode throughput + a small end-to-end GE run (criterion or a deterministic
  budget-based bench, no flaky wall-clock asserts in tests).
• Security/audit: cargo audit clean (or documented advisories).
• Docs/examples: crate-level rustdoc with a complete Target + Fitness example; ensure
  examples/evolve_toy.rs and examples/evolve_faber.rs build and run.
• Release: tag v0.1.0 (annotated). (PyPI/wheel is out of scope — no Python this phase.)

Acceptance criteria

• CI green on all jobs (fmt, clippy -D warnings, test --all-features, doc, feature matrix).
• cargo doc --no-deps builds with no warnings; rustdoc has a runnable end-to-end example.
• Benchmark suite runs (cargo bench or equivalent) and reports decode + GE-run numbers.
• cargo audit clean or advisories documented.
• Stability/semver doc present (grammar + snapshot formats + MSRV).
• v0.1.0 tag created.

Anti-goals

• Do not add PyO3 / wheel / PyPI (D7 — that's the later bridge phase).
• Do not add NSGA-II (D5) or resume/checkpoint (D4) — those are v2.
• Do not weaken the determinism contract or the validity-invariant tests to make CI pass.

Notes

Depends on M2 (#3) (and M3 (#4) for the final operator set). Completes the pure-Rust GGGP build:
M0 (#1) → M1 (#2) → M2 (#3) → M3 (#4) → M4. After this, a separate spec covers the Python bridge

• real Archimedes two-tier fitness (with DSR deflated for the GA's true trial count).