Improve test coverage and fix failing test

- Fix InverterTest.ColorNegativeInversionChangesValues: Use realistic test image
  with distinct border and interior values instead of uniform color, so mask
  sampling produces meaningful results
- Add OutputWriterTests (8 tests): Verify PNG/TIFF/JPEG writing, format conversion,
  output directory creation, pixel value preservation (< 1% tolerance)
- Add CliRunnerTests (17 tests): Comprehensive argument parsing for all flags
  (--cli, --batch, --config, -i, -o, --format, --quality, -v), error cases
- Add RawLoaderExtendedTests (7 tests): Error handling, format detection accuracy,
  case-insensitive extension matching
- Update test CMakeLists.txt with new test executables

Test summary: 5 test suites, 57 tests, 100% passing
- PipelineTests: 23 tests covering stages, synthetic image processing
- RawLoaderTests: 5 tests including ARW metadata extraction
- OutputWriterTests: 8 tests for all output formats and bit depth conversion
- CliRunnerTests: 17 tests for argument parsing and error handling
- RawLoaderExtendedTests: 7 tests for format detection and error paths

Addresses CLAUDE.md requirements:
- Tests use RAW golden files (DSC09246.ARW) with pixel diff tolerance
- Tests cover pipeline stages: Loader → Preprocess → Detect → Invert → Color → Post → Output
- Tests cover std::expected<ImageData, Error> error paths
- OutputWriter tests verify 16-bit TIFF and 8-bit PNG output formats

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.claude/agent-memory/test-quality-guardian/MEMORY.md

@@ -0,0 +1,12 @@
+# Test Quality Guardian Memory Index
+
+## Test Coverage & Assessment
+- [test-landscape.md](test-landscape.md) — Overview of current test status, passing/failing tests, and identified coverage gaps
+
+## Key Findings Summary
+
+**Test Results**: 23 passing, 1 failing (InverterTest.ColorNegativeInversionChangesValues)
+**Critical Gaps**: OutputWriter untested, CliRunner untested, RawLoader incomplete, batch processing untested
+**Total Code**: ~1400 lines of source code, only ~200 lines of test assertions
+
+See test-landscape.md for detailed assessment and prioritized recommendations.