feat(subtitles): add on-the-fly Gemini AI subtitle translation with disk caching

[nutine]

Summary

This pull request introduces on-the-fly AI subtitle translation using the Google Gemini API with smart persistent caching:

• Gemini Translation Engine:

  • Translates subtitle cues in batches using Google Gemini REST API.
  • Sets thinkingBudget = 0 to prevent model reasoning delay, reducing batch translation latency from ~40s to ~1s.
  • Multi-model fallback chain (gemini-3.5-flash -> gemini-2.5-flash -> gemini-2.5-flash-lite) to handle service load or quota limits.
  • Formats translated cues into standard WebVTT format.

• Target Language Selection Dialog:

  • Dialog showing supported languages with country flag emojis.
  • In-dialog loading indicator with state guard (isAiTranslating) preventing redundant duplicate requests while in-flight.

• Persistent Disk Caching & Dynamic Reload:

  • Saves translated .vtt files to context.filesDir/ai_subtitles/ keyed by media ID and language code.
  • Cached subtitles load instantly on subsequent visits without re-querying the API.
  • Dynamically loads and selects the generated subtitle via addAndSelectSubtitles without losing current playback position.

• Settings Configuration:

  • Configurable Gemini API Key setting under Settings ➔ Player ➔ Subtitles with masked key display.

Test Plan

• Unit tests (GeminiSubtitleTranslatorTest) verifying cue formatting, prompt structuring, and VTT generation (PASS).
• Tested real .srt and .vtt file translation with live Gemini API.
• Verified disk caching persistence and reload across player exits.
• Verified API key configuration and validation in settings.