Created a Complete onboard custom model training

Skipper can train a genuinely modified LLM model onboard from a selected custom dataset, register it as a versioned runtime model, restore the complete Assistant, preserve traceability, and automatically clean up unnecessary training storage.

Added TTS benchmark tools

Tools were added for Text To Speech benchmarking.

Centralized runtime config defaults

All config values are centralized for a more organized functionality.

Added AI model warmup

All used models during a session are now warmed up from start for faster responses.