perf: Add batched commits and LRU caching for database operations

[jamestexas]

Summary

Optimize database operations to significantly reduce execution time by eliminating per-record commits and adding query result caching.

Problem

Vunnel currently performs database commits after every single operation:

1. Fix-date tracking commits after every insert (~322k commits for full NVD run)
2. Result writes use separate transactions per CVE (~322k transactions)
3. Fix-date lookups hit the database every time, even for duplicate queries

This creates excessive disk I/O, especially when using NFS. This ultimately requires extra memory to process larger providers (like NVD) as the uncompressed provider database is ~15gb.

Solution

1. Batched Commits for Fix-Date Tracking

• Before: conn.commit() after every insert
• After: Batch commits every 2000 operations (configurable)
• Auto-flushes on context exit to ensure data integrity

2. Batched Commits for Result Writes

• Before: Separate transaction per CVE write
• After: Batch commits every 2000 operations (configurable)
• Maintains single active transaction, commits periodically

3. LRU Caching for Fix-Date Lookups

• Before: Every fixdater.best() call hit database
• After: functools.lru_cache(maxsize=10000) caches results
• Eliminates duplicate queries for same (CVE, CPE, version, ecosystem)
• Cache cleared on context exit

Performance Impact

Tested on NVD provider with 322k CVEs:

• Improvement: 35% faster (~6 minute savings on my mac).
• Database commits: Reduced from ~322k to ~161.

Testing

• All existing tests pass
• Added test for batched commit behavior
• Added test for cache hit functionality

Fixes: https://linear.app/chainguard/issue/PLA-368/optimize-vunnel-database-operations-with-batching-and-caching

Memory Safety

Batch size of 2000 ensures memory usage remains bounded while still providing significant performance gains.