Drop prepared statements for cached queries if connection gets swapped due to operations like ZDP

[JasonShin]

Cloud database providers like AWS Aurora offer Zero Downtime Patching (ZDP), which maintains connection availability during maintenance windows. However, ZDP has a critical limitation: prepared statements don't survive the underlying connection replacement that occurs during the patch process. This creates a mismatch between the client-side cache (node-mysql2's LRU cache of prepared statements) and the server-side reality. The client believes cached prepared statements are still valid, but the database has discarded them after the ZDP operation. When the application attempts to execute a cached prepared statement, it receives a "prepared statement not found" error.

Proposed Solution Implement resilient prepared statement handling that can recover from ZDP scenarios:

• Error Detection: Catch prepared statement errors that indicate the statement no longer exists on the server
• Cache Invalidation: Remove the invalid entry from the LRU cache
• Transparent Recovery: Automatically re-prepare the statement and retry the original query
• Seamless Experience: Handle this recovery transparently so the application layer doesn't need to manage ZDP-related failures

This approach would make the prepared statement cache self-healing, automatically adapting when cloud infrastructure operations invalidate server-side prepared statements while maintaining client-side cache consistency.

The key insight is treating prepared statement "not found" errors as cache invalidation signals rather than terminal failures, enabling graceful recovery from infrastructure-level operations like ZDP.