refactor: move video processing configuration into nested processor_kwargs for improved parameter organization

feat: implement PyAV-based video frame extraction and update model processing parameters for MiniCPM-V 4.6

refactor: simplify file type detection logic and wrap processor template input preparation in torch.no_grad

feat: add file extension fallback for mime type detection to support diverse media inputs

refactor: remove PyAV dependency and update media processing to use file paths directly

perf: enable multi-threaded video decoding and offload file persistence to background thread

feat: implement file persistence for audit logging, improve text normalization logic, add beam search support, and update UI with a few-shot builder interface.

chore: silence asyncio logging and remove default greeting from chat interface

refactor: optimize video frame extraction by implementing direct seek-based sampling with duration-based fallback

feat: implement fallback to video loading if image processing fails in file handling logic

fix: add exception logging for HTTP requests and rename server instance to demo

feat: implement HTTP request logging middleware and add tabbed navigation interface

feat: implement client ID tracking, session logging, and enhanced message history processing

feat: implement multi-modal chat history support, regenerate functionality, and few-shot example capabilities

feat: add Generator return type hint to streaming inference endpoint

feat: update processor_kwargs to include frame sampling configuration for videos

feat: implement streaming inference with history support and update UI theme

fix: add do_sample_frames flag to processor_kwargs to prevent metadata errors

refactor: add video loading error handling and update processor_kwargs for modern transformers compatibility

feat: implement manual video frame loading using PyAV to support direct frame passing for video processing

feat: add return type annotation to predict function in app.py

fix: update device mapping and GPU duration, and add thinking UI state to chat interface

feat: add text-only inference support and conditional downsample_mode parameter for model generation

feat: implement multimodal MiniCPM-V 4.6 inference application with FastAPI and custom web interface