Complete V3 row ID assignment and inheritance

[osipovartem]

Summary

• complete Iceberg v3 row-ID assignment and inheritance across V1, V2, and V3 manifest-list upgrades
• decode historical manifest lists from their embedded writer schema
• hydrate counters missing from V1 manifests before V2/V3 serialization
• preserve partition summaries across schema/spec evolution, missing snapshot schema_id, and expired historical snapshots
• restrict metadata promotion to Iceberg-legal integer, floating-point, and same-scale decimal widening
• retain compatible lower/upper bounds while discarding incompatible optional metrics
• route partition INT -> LONG and FLOAT -> DOUBLE evolution through the explicit Iceberg-promotion path while keeping generic casts strict

Performance

• V1 counter hydration and manifest inspection use bounded concurrency of eight tasks
• candidate partition decoders are cached by partition-spec and preferred-schema identity
• unchanged values use equality/no-allocation fast paths; conversion occurs only for actual widening
• no per-row evaluation or change to the vectorized Arrow/DataFusion scan path is introduced

Validation

• manifest-list specification tests pass, including V1-to-V2 and V1-to-V3 upgrade coverage
• all 172 active iceberg-rust-spec tests pass
• all 10 transaction operation tests pass
• focused bounds-promotion and historical-schema/spec regressions pass
• cargo +1.94.0 clippy -p iceberg-rust -p iceberg-rust-spec --all-targets -- -D warnings
• git diff --check
• independent read-only review covering correctness, regressions, performance, API compatibility, and missing tests: APPROVED

The implementation preserves the existing DF55 manifest cache and vectorized scan path.