fix(transactions): improve recovery liveness and diagnostics
Depends on #10378 and should be rebased/merged after it. The expected rebase resolution in TransactionQueue.AbortAndRestoreCore preserves #10378's discarded-batch completion around #10374's correlated restore/diagnostics; TransactionQueueStorageWorkTests also needs the new diagnostic identity constructor argument. This combined resolution has been validated on net8.0 and net10.0.
Transaction recovery can linger after orphaned managers, while the existing diagnostics do not expose enough activation, role, phase, lock, restore, and cancellation state to explain liveness failures.
This change adds provider-neutral transaction recovery diagnostics, adaptive recovery-ping retries, and bounded failure detection with explicit outcomes. Local transaction-manager aborts dispatch every target in exactly one manager-owned cancel fan-out before publishing the durable abort decision, preventing TransactionAgent duplication and avoiding the caller response deadline depending on remote cleanup. The already-dispatched cleanup remains observed and is bounded locally by activation cancellation and the configured lock-timeout budget while retaining the original abort status.
Storage-write diagnostics carry the exact committed transaction IDs, so durable local-commit cleanup gates can only be released by confirmation of the gated transaction rather than an unrelated recovery probe. Deterministic phase gates cover manager wait, persisted remote prepare, durable local commit before matching confirmation/collection, and activation cancellation during blocked cancel fan-out. Azure Storage and DynamoDB suites include the provider-backed recovery cases.