fix(player): fall back to MPV playable source when watching anime and classify sentence audio export failures

[Jolimunium]

Problem

When watching anime in the player, exporting sentence audio during scene mining could fail silently or produce generic error toasts without actionable feedback. Typical failure causes included:

• Invalid MPV track index mappings when MPV's frozen ff-index was missing or mismatched relative to FFmpeg streams.
• Unreadable or unseekable original video sources (e.g., remote HTTP/stream read errors, DRM, or protected streams).
• Restricted or unwhitelisted audio codecs.
• Lack of fallback to MPV's playable video source when the original video source encountered audio extraction issues.
• Generic Anki media warnings that did not specify the root cause of sentence audio generation failures.

Changes

1. Sentence Audio Fallback & Track Resolution (SceneSentenceAudioService.kt, SceneVideoInput.kt, SceneCaptureRequest.kt)

• Added SceneVideoInputOrigin (ORIGINAL_VIDEO, PLAYABLE_VIDEO, EXTERNAL_AUDIO) to track source provenance.
• Introduced ScenePlayableFallbackResolution and resolvePlayableFallback() to safely verify and resolve MPV's playable video source when the original video source fails.
• Refactored SceneSentenceAudioService with multi-stage probing (Selected Track Probe → All Stream Probe → Unrestricted Discovery Probe) and automatic fallback to the playable source for retryable errors (isPlayableFallbackRetryable).
• Implemented automatic track index recovery when exactly one readable audio stream exists, even if MPV's ff-index property is unpopulated.

2. Failure Classification & Safe Native Diagnostics (FfmpegKitSceneCommandExecutor.kt, SceneMediaProbe.kt, SentenceAudioDiagnosticLog.kt)

• Added SceneFfmpegFailure enum and classifySceneFfmpegFailure() to categorize native FFmpeg/FFprobe failures (STREAM_MAPPING, SOURCE_READ, SEEK, OUTPUT_WRITE, UNKNOWN).
• Enhanced SceneMediaProbe with inspectSelectedAudio() and audioStreams() to detect protection markers, codec types, and stream blocks.
• Implemented SentenceAudioDiagnosticJournal and StorageFolderSentenceAudioDiagnosticLogger to record diagnostic events in storage while automatically redacting sensitive data (URLs, authorization tokens, headers, and local file paths).

3. Detailed UI Warnings & Translations (AnkiMedia.kt, AnkiCardCreator.kt, PlayerSceneMiningUi.kt, strings.xml, PlayerViewModel.kt)

• Defined AnkiSentenceAudioFailure, AnkiSentenceAudioDiagnostic, and AnkiSentenceAudioPreparation in the Anki module.
• Added AnkiMediaWarning.SentenceAudioGenerationFailed and SentenceAudioStorageFailed.
• Updated PlayerSceneMiningUi.kt and added 27+ localized string resources in strings.xml to present detailed diagnostic toasts specifying whether issues arose from original video sources, MPV playable fallback sources, or external audio tracks.

4. Comprehensive Unit Tests

• Added and updated tests in eu.kanade.tachiyomi.ui.player.scene:
  • SceneSentenceAudioServiceTest.kt: Playable fallback, track index recovery, probe failures, codec restrictions, external audio, and timeouts.
  • SceneCaptureRequestTest.kt: Subtitle capture request resolution and single-audio tracking.
  • SceneVideoInputTest.kt : Playable fallback resolution and discovery probe arguments.
  • SentenceAudioDiagnosticLogTest.kt : Log redaction (URL/token/header masking) and byte budget retention.
  • SceneFfmpegFailureClassifierTest.kt : FFmpeg failure regex classification.

Result

• When exporting sentence audio while watching anime, if the original video source fails, the system automatically falls back to MPV's playable video source to successfully generate sentence audio.
• Users receive precise, localized toast warnings detailing the exact failure reason instead of generic errors.
• Sensitive credentials and stream URLs in native diagnostic logs are automatically redacted before storage.

[sohilsayed]

the animated mining will be reworked so if you can fix the conflicts that would be good

[Jolimunium]

I've updated this in #106 instead.