feat(ant-node): persist record index to eliminate startup CPU spike

[tomdif]

Summary

• Cache the record metadata index to disk so node restarts read one small msgpack file instead of doing a parallel AES-256-GCM-SIV decrypt of every record (up to 16K files)
• Eliminates the 40+ second 100% CPU spike on startup after binary upgrades (same peer ID, same network ID)
• Caps rayon threads in the full-scan fallback to available_parallelism - 1, reserving one core for the OS/tokio runtime

How it works

1. On startup, try_load_cached_index() reads a small record_index msgpack file from historic_quote_dir
2. Validates the encryption seed matches (detects peer/encryption changes) and spot-checks 10 files on disk
3. If valid, skips the expensive update_records_from_an_existing_store() full-decrypt scan entirely
4. If invalid or missing, falls back to the existing full scan (now thread-capped)
5. Index is flushed on payment, on initial construction, and synchronously on Drop

Security

• Records stay AES-256-GCM-SIV encrypted on disk — unchanged
• Index contains only metadata already visible as plaintext filenames + enum tags
• Encryption seed stored in index — mismatches trigger automatic full rescan
• Tampered index worst case: node tries to serve a record that doesn't exist → fails → network re-replicates

Test plan

• cargo check -p ant-node — passes
• cargo clippy -p ant-node --all-targets -- -Dwarnings — passes clean
• cargo test --release --package ant-node --lib — 76/76 tests pass (including can_store_after_restart)