Expose NOCR/BinaryOCR database path through seconv MCP

[Blackspirits]

Summary

Upstream PR SubtitleEdit#14129 exposes convert_subtitle.ocrEngine with nocr and binaryocr as supported values, but the MCP tool does not expose the database path required by either engine.

The shared OCR factory requires:

• nocr → ConversionOptions.OcrDb pointing to a .nocr database;
• binaryocr → the same option pointing to a .db database.

Because SubtitleEdit#14129 leaves OcrDb null unconditionally, those two advertised MCP engine choices cannot succeed on image-based input.

This change:

• adds optional ocrDb to convert_subtitle;
• documents that it is required for NOCR/BinaryOCR and which extensions they use;
• passes the value through to ConversionOptions.OcrDb;
• leaves Tesseract/Ollama/Paddle/llama.cpp behavior unchanged.

Validation

• based exactly on upstream PR seconv: add seconv mcp, a Model Context Protocol server over stdio SubtitleEdit/subtitleedit#14129 HEAD ad51c0ae9f5e8942f281e4acd82090974f9b8847
• one MCP adapter file only
• preserves the existing CLI/core OCR contract rather than adding a new one
• full solution CI pending

AI assistance: ChatGPT was used to audit behavioral parity between the new MCP tool surface and seconv's existing CLI/core OCR options.