Add halogen-image module for image-to-theme extraction

[himattm]

Summary

• New halogen-image KMP module that extracts dominant colors from images and maps them to Halogen themes
• Uses HCT-space k-means clustering for perceptually accurate color extraction (~0.8ms regardless of image size)
• Coil 3 integration for URL-based image loading with platform-specific pixel extraction (Android/JVM/iOS/wasmJs)
• Two paths: algorithmic (toSpec() — instant, no LLM) and LLM-enhanced (toHint() — mood-aware typography/shapes)

API

// Music player — one line, URL is the cache key
engine.resolveImage(album.artUrl, imageLoader, context)

// Raw pixels (for non-Coil loaders)
engine.resolveImage("album:123", pixels, width, height)

// Standalone extraction
val colors = ImageQuantizer.extract(pixels, w, h)
colors.toSpec()   // algorithmic
colors.toHint()   // enriched LLM hint

Module structure

• ImageQuantizer — histogram + HCT k-means clustering (subsamples to 16K pixels, 64 histogram entries, 10 iterations)
• ImageThemeExtractor — assigns extracted colors to pri/sec/ter/neutral roles by chroma/hue/tone heuristics
• DominantColors — result type with toSpec() and toHint() conversions
• EngineExtensions — resolveImage() overloads on HalogenEngine
• Platform expect/actual for Coil image → ARGB pixel extraction

Performance

Image size	Extraction time
500×500	0.80ms
1080×1080	0.86ms

Tests

38 test cases across 3 files covering extraction algorithm, role assignment heuristics, round-trip validation through ThemeExpander.expand().

Test plan

• ./gradlew :halogen-image:jvmTest — all 38 tests pass
• ./gradlew :halogen-image:testReleaseUnitTest — Android tests pass
• ./gradlew :halogen-image:apiCheck — API surface validated
• Existing module tests unaffected (halogen-core, halogen-engine)
• wasmJs browser tests have pre-existing Skiko resolution issue (same as halogen-compose)

🤖 Generated with Claude Code