Expand primitive/literal corpus and OXC mapping evidence

[swernerx]

Summary

• expand the shared primitive/literal corpus across declarations, expressions, widening, as const, satisfies, contextual literals, enum-like values, imports, unions, null-like values, and response budgets
• require structured Go-oracle expectations and one explicit Rust classification for every selected fact, with no hidden denominator exclusions
• classify the three recovery parser gaps individually as stable expected oxc-parse-error mappings and keep unexplained semantic, transport, or mapping differences blocking
• run the complete conformance report twice and require byte-identical output

Go remains the semantic authority and production fallback. This does not change the TS7 producer protocol, external consumers, dependencies, or Palamedes.

Evidence

The expanded report passes with 28 classified facts: 20 supported, 4 expected unsupported, 1 expected budget, and 3 expected mapping observations. All 29 supported records match, all 100 roots retain exact identity agreement, and there are no unexplained differences.

Local validation

• RUSTUP_TOOLCHAIN=1.97.1 cargo fmt --manifest-path internal/oxc_reference/Cargo.toml --all -- --check
• RUSTUP_TOOLCHAIN=1.97.1 cargo clippy --locked --manifest-path internal/oxc_reference/Cargo.toml --all-targets -- -D warnings
• RUSTUP_TOOLCHAIN=1.97.1 cargo test --locked --manifest-path internal/oxc_reference/Cargo.toml
• RUSTUP_TOOLCHAIN=1.97.1 ./internal/oxc_reference/run-conformance.sh --output /tmp/issue46-report-final.json
• go test -count=1 ./internal/semanticfacts ./internal/tsfacts ./cmd/tsfacts
• go build -tags=noembed ./cmd/tsfacts
• npx hereby check:format
• npx hereby lint
• npx hereby test

AI disclosure

This pull request was implemented with OpenAI Codex assistance and fully reviewed and validated locally before publication.

Closes #46