Add model response output to GitHub job summary and support multiple model presets

[kirklandsign]

Summary

• Add model response extraction from instrumentation tests to GitHub job summary
• Support configurable model presets (stories, llama, qwen3, custom) with proper filename handling
• Pass model filenames to instrumentation tests via Gradle arguments for preset-aware testing

Changes

GitHub Workflow

• Extract LLM conversation output from logcat and display in job summary
• Add debug output showing which model preset and files are being used

Gradle Configuration

• Keep original model filenames when pushing to device (instead of renaming to model.pte)
• Pass modelFile and tokenizerFile to instrumentation tests via runner arguments
• Add skipModelDownload flag for local testing with existing files

Instrumentation Tests

• SanityCheck and UIWorkflowTest now read model filenames from instrumentation arguments
• Default to stories preset (stories110M.pte, tokenizer.model) for Android Studio compatibility
• Log model responses to logcat with LLAMA_RESPONSE tag for CI extraction

Documentation

• Add README section documenting model preset usage and available options

Test plan

• Run ./gradlew connectedCheck -PmodelPreset=stories locally
• Run ./gradlew connectedCheck -PmodelPreset=llama and verify different model is used
• Trigger CI workflow and verify model response appears in job summary
• Run tests from Android Studio play button (should use default stories preset)