feat(executions): add System Error status for SQS and infrastructure failures

[joelorzet]

Summary

Adds a distinct system_error execution status, persisted wherever a run fails for a platform/infrastructure reason (error_type = system) rather than a user/workflow error. System failures, including SQS issues, are now visible and filterable apart from user errors, end to end.

This builds on the existing system-vs-user error model (KEEP-693 / KEEP-545 / TECH-6544); it does not replace error_type/error_category/error_code, it adds a status that mirrors them so operators get a status they can see and filter.

Both SQS failure directions are covered

• Producer enqueue fails (schedule/block/event SendMessage throws): the pre-created phantom row resolves to system_error (CS-0001/BS-0001/ES-0001/N-0002).
• Consumer never receives (message lost / dispatcher down): the reaper ages the phantom to system_error (P-0005, reclassified from workflow_engine to infrastructure).
• Consumer fails mid-process: a new backstop marks the in-flight run system_error (E-0004) and deletes the message immediately, instead of leaving it for the reaper (up to 30 min later). A typed RequeueSignal preserves the deliberate concurrency re-queue so it is never marked or duplicated.
• Dispatch / k8s / runtime failures: map to system_error via P-0002/P-0004 and the classifier on the completion path.

How it stays consistent

A single rule, statusForErrorType(errorType), drives every write site. Every error-count query and terminal-state check matches both error and system_error, so no metric or wait path silently drops system errors:

• getSystemErrorsByCategoryFromDb (the platform gauge the infra alert reads), the managed-org per-workflow gauge, the per-status execution gauge (gains a status=system_error series), and the duration histogram
• analytics error buckets, x402 isTerminalStatus, the self-heal CAS, getCustomerRunErrorMessage, and the failure digest

UI

Distinct amber "System Error" badge and filter in the runs table, and a dedicated label/icon/dot in workflow runs.

Design note for review

The consumer backstop deletes the SQS message after marking system_error. This prevents poison-message redelivery loops and duplicate execution rows. Trade-off: a transient consumer error no longer auto-retries via SQS redelivery; it surfaces immediately as a system error and the run remains re-triggerable. The concurrency back-pressure path is unaffected (it re-queues as before). Worth a careful look during review.

No DB migration: workflow_executions.status is a plain text column with no check constraint.

Validation

• pnpm type-check: clean
• pnpm check: changed files clean
• Unit suite: 6451 passed, 1 skipped (incl. a new execution-status test)
• DB e2e: reaper-codes, executor-status-backstop pass; scheduler and event-tracker package suites pass
• Updated expectations: reaper, both phantom suites, and two infra e2e assertions (SIGTERM and RPC failures now classify as system)