[FEATURE] #880 - Unit and integration tests for batch-level learning and strategy memory

[stephanj]

Parent Issue

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

Task

Comprehensive test coverage for the batch-level learning and strategy memory feature.

Test Cases

TaskExecutionMetrics Tests

• Metrics captured for successful task (timing, tokens, cost)
• Metrics captured for skipped task (reason, zero tokens)
• Metrics captured for failed task (error summary, exit code)
• Timer accurate within reasonable tolerance (±100ms)
• CLI mode captures execution time from process lifecycle
• LLM mode extracts token usage from ChatMessageContext

BatchExecutionMemory Tests

• Record and retrieve single task learning entry
• Retrieve entries for specific dependency task IDs
• getSummarizedContext() respects max character limit
• Older entries compressed to one-liners
• Recent entries include full details
• Failed tasks always included regardless of age
• clear() empties all entries
• Thread-safe concurrent record/read

Context Injection Tests

• Learning context injected when LearningMode is BATCH
• No learning context when LearningMode is OFF
• Context within character budget
• Pattern observations included in context
• Both CLI and LLM modes receive learning context

BatchRunSummary Tests

• Summary aggregates all per-task metrics correctly
• Status breakdown counts match individual task statuses
• Total tokens = sum of per-task tokens
• Total cost = sum of per-task costs
• Console output formatted correctly

ExecutionPatternExtractor Tests

• COMMON_ERROR detected when 2+ tasks fail with same error
• FILE_HOTSPOT detected when 3+ tasks modify same file
• PERFORMANCE_TREND detected when tasks progressively slow down
• No patterns emitted for single-task run
• Patterns are concise and under 200 chars each
• Confidence values are reasonable (0.0-1.0)

Integration Tests

• Full batch run with LearningMode=BATCH: each task receives prior context
• Full batch run with LearningMode=OFF: no cross-task context (regression)
• 5-task batch: verify growing learning context per task
• Learning context does not exceed max limit even with many tasks

Files to Create

• src/test/java/com/devoxx/genie/model/spec/TaskExecutionMetricsTest.java (new)
• src/test/java/com/devoxx/genie/service/spec/BatchExecutionMemoryTest.java (new)
• src/test/java/com/devoxx/genie/service/spec/ExecutionPatternExtractorTest.java (new)
• src/test/java/com/devoxx/genie/service/spec/BatchLearningIntegrationTest.java (new)

Dependencies

• Requires: All other [FEATURE] Introduce Batch-Level Learning and Strategy Memory #880 sub-tasks

Acceptance Criteria

• All test cases above pass
• No flaky tests (mock timers, use deterministic data)
• Existing tests unaffected
• OFF mode regression tests confirm zero behavior change