Optimize Indexes

[kraftp]

Optimize workflow_status table indexes to reduce the cost of maintaining indexes at scale.

• Indexes on parent_workflow_id, forked_from, and deduplication_id are made partial on those fields not being NULL so they're not maintained when not needed.
• The index on executor_id is dropped as use in memory map to enforce worker concurrency #644 eliminates the need for it.
• The index on the status field is split into two partial indexes: one on PENDING workflows for use in recovery; the other on failed workflows (ERROR, CANCELLED, MAX_RECOVERY_ATTEMPTS_EXCEEDED) for rapid troubleshooting. This reduces the number of times the index is updated during a typical workflow lifetime (ENQUEUED->PENDING->SUCCESS).
• The index on queue_name and status used by the main dequeue query is optimized to index on ("queue_name", "status", "priority", "created_at"), the exact predicates and sort order of the dequeue query. It is now partial on ENQUEUED or PENDING workflows as only the dequeue query needs it.
• A new partial index is created for rate-limited queues to quickly find recent queries in that queue. That index is only maintained for rate-limited queues (using a new rate_limited field).
• All migrations use CREATE INDEX CONCURRENTLY for safety when running against production workloads.