Revalidate unique-index point reads

[pathscale]

Summary

• revalidate the returned row field after resolving a unique-index link
• cover in-memory, persisted, and read-only generated table families
• preserve OrderedFloat equality semantics for floating-point unique indexes
• add deterministic stale-link regressions for integer and float keys

Correctness boundary

A stale unique-index entry can no longer make select_by_<unique> return a row whose indexed field differs from the requested key. A mismatch returns None. This change does not claim that the lookup is linearizable or retry a concurrently changing index.

Verification

• cargo test --workspace --all-targets: passed (486 passed, 5 ignored across the main suites, plus workspace crate tests)
• cargo fmt --all -- --check: passed
• cargo clippy --all-targets -- -D warnings: passed
• cargo clippy --workspace --all-targets -- -D warnings: reaches a pre-existing clippy::useless_conversion failure in performance_measurement/codegen/src/performance_measurement.rs:90

Performance

Criterion unique_index_select_by_test, seven alternating before/after pairs (30 samples, 0.5 s warm-up, 1 s measurement):

• master mean: 56.907 ns
• patched mean: 56.533 ns
• delta: -0.66%
• patched result was faster in 6 of 7 pairs

No measurable regression was found, so this correctness check is not feature-gated.

[pathscale]

Superseded by #187, which consolidated and merged the backend-selection and publication hardening work after the combined review and validation rounds.