Improve Tron restart times by adding parallelization to job_collection.restore_state()

[CharlesPeacock]

Follow-up to the series of Tron restart improvements.

As of recently, Tron has been seeing restart times of over 6 minutes in pnw-prod. Optimizing these restart times allows for less downtime and more opportunities to experiment with new changes.

This change adds a ThreadPoolExecutor to parallelize job_collection.restore_state(). By setting max_workers to 5 we can see a significant improvement in state restoration times:

Original performance (tested with copy of tron-pnw-devc dynamodb table):

Run: linear run (Tue Jul 28 10:32:01 AM PDT 2026)
Total startup: 93.59s
Schedulers built: 3.786149263381958s
DynamoDB retrieval: 10:27:25 -> 10:27:55 (30.000s)
Apply state: 10:27:55 -> 10:28:53 (58.000s)
Unprocessed key retries: 5
Log: /tmp/output_log_linear_attempt1.log
Restore state: 57.425s <----- 

Performance after parallelization (same environment as original test, 5 workers):

Run: parallel run (Tue Jul 28 10:19:16 AM PDT 2026)
Total startup: 66.10s
Schedulers built: 3.781592607498169s
DynamoDB retrieval: 10:14:21 -> 10:14:52 (31.000s)
Apply state: 10:14:52 -> 10:15:22 (30.000s)
Unprocessed key retries: 7
Log: /tmp/output_log_parallel_attempt2.log
Restore state: 29.658s <-----

Tests passed