feat(reliability): wire FailurePatternDetector — learn on workflow failure + detect/flag recurring patterns #10628

Subtask of #10602 (umbrella #10603). The FailurePatternDetector (services/failure_pattern_detector.py) is fully built with a singleton (get_pattern_detector()) but has zero production callers on either side — nothing writes patterns and nothing reads them.

Scope (full loop — write + read consumer)

• Write: on workflow execution failure (enhanced_orchestration/workflow_runner.py:_handle_workflow_execution_failure), call learn_pattern(causal_chain, error_type) to record/increment the failure pattern (Redis-backed). Fire-and-forget, never raise into the failure path.
• Read: before/at failure handling, call detect_pattern(causal_chain, error_type); when a known recurring pattern is found, annotate the failure result (e.g. known_failure_pattern: {occurrences, resolution_success_rate}) so it's surfaced to the caller/logs — this is the consumer that makes the learned data useful.
• Guard all detector calls so a Redis/detector error never breaks workflow failure handling.

Verification

• Unit test: two identical failures → learn_pattern increments occurrence count; detect_pattern returns the known pattern and the failure result carries the known_failure_pattern annotation on the second occurrence.