feat(perf): New indexes for job pickup & monitoring

[smudge]

This PR is the first of several query optimizations, and can largely be read in two parts:

1. A pair of heavily-annotated migrations that add new index(es) as safely as I could muster generically (w/ timeouts and a lock retry loop). I also updated the README w/ some expanded instructions and ⚠️ safety warnings. ⚠️
2. The resulting changes to all of the EXPLAIN snapshots, which seems to improve numerous plans without regressing any. (I haven't yet removed the old index, of course, but that's the eventual goal, once I make a few more query changes.)

(Non-AI) Summary of EXPLAIN changes:

• worker
  • postgres: failed_at IS NULL is no loner necessary in the Filter:, and when a single queue is specified, it gets promoted to an Index Cond:
  • sqlite: SCAN ... USING INDEX either way (doesn't reveal more than that)
  • mysql: Table scan becomes Index lookup 💪 with both failed_at and queue in the index conditions! (A Sort is also completely avoided as it's able to use the pre-sorted index now.)
• monitor
  • postgres:
    • again, failed_at is no longer necessary in Filter: clauses.
    • failed_count goes from Seq Scan to Index Only Scan 💯
    • max_lock_age and working_count go from Seq Scan to Index Scan 👍
  • sqlite:
    • for several (but not all) metrics, SCAN delayed_jobs ... becomes SCAN delayed_jobs USING INDEX ...
  • mysql:
    • count and future_count go from Table scan to Covering index scan
    • failed_count goes from Table scan to Covering index range scan
    • max_lock_age, max_age, working_count, workable_count, and alert_age_percent go from Table scan to Index lookup w/ index conditions

/no-platform

[effron]

One question about the migration! Generally, is there a chance this impacts write throughput? should we prioritize removing the old indices if they are not needed anymore?

[effron]

domainLGTM!