Introduces a new LLM (GPT) fallback mechanism for series matching when traditional parsing fails. Adds ILlmSeriesMatchingService with OpenAI, caching, and rate-limiting implementations. Integrates LLM matching into ParsingService, adds new config options, and exposes a REST API for configuration and testing. Includes unit and integration tests, and updates DI and solution files. This improves matching for ambiguous, foreign, or scene releases in a safe and configurable way.

Description

This PR adds an optional LLM-based fallback matching system for series identification when traditional parsing methods fail. The feature uses OpenAI-compatible APIs to intelligently match release titles to series in the user's library, significantly reducing manual import interventions.

Key Features:

• Fallback-only: LLM matching only triggers after all traditional matching methods (scene mapping, TVDB/IMDb/TVRage ID, title-based, alternate titles) have failed
• Confidence threshold: Configurable threshold (default 70%) - matches below threshold require manual confirmation
• Cost control: Built-in rate limiting (default 60 calls/hour) and response caching (24h) to minimize API costs
• Encoding resilience: Handles miscoded umlauts (e.g., ü → ü), foreign scripts (Japanese, Chinese, Korean, Cyrillic, Arabic), and various encoding issues common in release titles
• Anime support: Recognizes alternate/localized titles (e.g., "Shingeki no Kyojin" → "Attack on Titan")
• Provider flexibility: Supports OpenAI, Azure OpenAI, and local LLM endpoints (Ollama, LM Studio)

Architecture:

• Decorator pattern: Rate Limiting → Caching → OpenAI API
• New SeriesMatchType.Llm for tracking match source
• Fully optional - disabled by default, requires API key configuration

Database Migration

YES - Adds the following configuration properties:

• LlmMatchingEnabled (bool, default: false)
• OpenAiApiKey (string, encrypted)
• OpenAiApiEndpoint (string, default: OpenAI API)
• OpenAiModel (string, default: gpt-4o-mini)
• LlmConfidenceThreshold (double, default: 0.7)
• LlmMaxCallsPerHour (int, default: 60)
• LlmCacheEnabled (bool, default: true)
• LlmCacheDurationHours (int, default: 24)