[FEATURE] #880 - Implement ExecutionPatternExtractor for cross-task pattern detection

[stephanj]

Parent Issue

Part of #880 — Introduce Batch-Level Learning and Strategy Memory

Context

Even with per-task learning entries, there is no mechanism to detect patterns across tasks — e.g., "all tasks touching the service/ directory take 2x longer" or "tasks that fail on timeout usually need to be split". A pattern extractor would analyze completed task telemetry and produce actionable observations.

Task

Create an ExecutionPatternExtractor that identifies recurring patterns from batch execution data.

Key Changes

1. Create ExecutionPatternExtractor:

   public class ExecutionPatternExtractor {
       /** Analyze completed metrics and extract patterns */
       List<ExecutionPattern> extract(List<TaskExecutionMetrics> metrics,
                                      List<TaskLearningEntry> entries);
   }

2. Create ExecutionPattern model:

   @Data
   public class ExecutionPattern {
       PatternType type;
       String description;      // human-readable observation
       double confidence;       // 0.0-1.0
   }
   
   enum PatternType {
       COMMON_ERROR,            // same error repeated across tasks
       PERFORMANCE_TREND,       // tasks getting slower/faster
       FILE_HOTSPOT,            // same files modified repeatedly
       STRATEGY_SUCCESS,        // specific approach works well
       STRATEGY_FAILURE         // specific approach keeps failing
   }

3. Pattern detection rules (heuristic-based, no ML):

   • COMMON_ERROR: If 2+ tasks fail with same error pattern → note it
   • FILE_HOTSPOT: If 3+ tasks modify same file → flag potential merge conflict risk
   • PERFORMANCE_TREND: If tasks are taking progressively longer → flag context window pressure
   • STRATEGY_SUCCESS: If tasks with web search enabled succeed more → recommend it
   • STRATEGY_FAILURE: If timeout failures cluster on large tasks → suggest splitting

4. Inject patterns into BatchExecutionMemory:

   • After each task completes, re-run extractor on accumulated data
   • Include pattern observations in learning context (via SpecContextBuilder)

Files to Create

• src/main/java/com/devoxx/genie/service/spec/ExecutionPatternExtractor.java (new)
• src/main/java/com/devoxx/genie/model/spec/ExecutionPattern.java (new)

Dependencies

• Requires: TaskExecutionMetrics ([FEATURE] #880 - Create TaskExecutionMetrics model for per-task telemetry #910)
• Requires: BatchExecutionMemory ([FEATURE] #880 - Create BatchExecutionMemory service for cross-task learning summaries #911)

Acceptance Criteria

• Pattern extractor identifies common errors across tasks
• File hotspots detected when multiple tasks touch same files
• Performance trends flagged (getting slower/faster)
• Patterns are concise (1-2 sentences each)
• Pattern list injected into learning context
• No patterns emitted for single-task runs (insufficient data)