fix: batch schema markers during shadow SQLite setup

[johoja12]

Problem

Each startup rebuilds two shadow SQLite databases. The entrypoint copied each of 447 schema_migrations markers by spawning a separate sqlite3 process, once for each database. On our QNAP canary this added several minutes to startup and made crash/restart diagnosis slow.

Fix

Generate escaped INSERT OR IGNORE statements from PostgreSQL and run them in one SQLite transaction per shadow database. Fail if either database command fails or the final row count differs from PostgreSQL.

Verification

The isolated QNAP trial used this entrypoint on a fresh import and again on a container recreation. Both shadow databases reported all 447 migration markers and Plex reached HTTP 200. The production Plex container remains on SQLite.