Retry queued transactions after transient connection loss

[Minecraft0122]

Summary

• keep the current transaction at the head of Plan's single-threaded database queue when a transient/recoverable connection failure occurs
• retry with bounded exponential backoff from 1 to 30 seconds so later transactions cannot overtake failed work
• stop retrying promptly when the database is closing and preserve interruption status
• recognize JDBC transient/recoverable exceptions and SQLState class 08, while excluding fatal and explicitly non-transient failures
• replace recursive invalid-connection lookup in MySQLDB with a transient exception handled by the queue retry path

The buffer is intentionally in-memory for the lifetime of the running server process; this PR does not claim crash/restart persistence.

Tests

• verifies transient exceptions through JDBC type and SQLState cause chains
• verifies statement, fatal, and non-transient failures are not retried
• verifies an invalid MySQL connection is closed and rejected after one pool lookup rather than recursing
• verifies a temporarily failing transaction completes before the transaction queued behind it and the queue drains to zero
• targeted tests and :common:checkstyleMain / :common:checkstyleTest pass locally

This is the temporary-database-outage portion of the closed #4772, split into a single-purpose PR as requested.

Closes #4727