Use checkpoint offset for timestamp startup restore

For startup.mode = timestamp, SeaTunnel could recompute the original configured timestamp after a checkpoint restore instead of honoring the concrete binlog file and position already persisted by the reader. In long-running jobs, that could move recovery back to an old timestamp anchor and fail when older binlogs were purged or corrupted.

Added a timestamp-only offset guard, reused restored binlog offsets directly during startup offset resolution, skipped the timestamp filter for restored concrete offsets, and pinned the behavior with focused MySQL CDC unit tests.

• Improves reliability for long-running MySQL CDC pipelines that use timestamp startup and checkpoint recovery.
• Prevents restored jobs from rewinding to the original timestamp when a checkpoint already contains a concrete binlog position.
• Avoids unnecessary timestamp-to-binlog lookup and timestamp filtering on the restore path.

• Added unit coverage for timestamp-only bootstrap offsets, restored binlog offsets, and binlog timestamp-filter decisions.
• Verified with the connector Maven package command documented in the PR.
• Merged after maintainer approval on July 25, 2026, closing the linked MySQL CDC recovery issue.