docs fixes: Add readme and architecture.

docs: Add readme and architecture.

docs: Add project roadmap (Excalidraw), results summary, and training/validation/test loss and metric curves.

feat(endpoint deployment): Add secure SageMaker endpoint deployment script using environment variables for IAM role, S3 model URI, and endpoint name.

feat(launcher script): Add SageMaker training launcher script with TensorBoard logging, hyperparam config, and S3 data channels.

feat(predict funtion inference): Add SageMaker predict_fn for multimodal video segmentation, inference, and top-3 emotion/sentiment predictions with Whisper ASR.

feat(model function inference): Add SageMaker model_fn for inference with FFmpeg check, multimodal model/transcriber setup, and device-aware loading.

feat(i/o inference sagemaker): Add SageMaker-compatible input/output handlers for S3 video processing and JSON prediction serialization.

feat(s3 utility inference): Add utility to download video files from S3 URI to a local temporary file for inference.

feat: Add VideoUtteranceProcessor for precise video segment extraction with FFmpeg, supporting utterance-level processing for multimodal models.

feat(deployment inference Audio Processor): Add AudioProcessor with FFmpeg-based audio extraction, Mel spectrogram conversion, normalization, and fixed-length padding/truncation for inference.

feat(deployment inference video processor): Add VideoProcessor class for video frame extraction, resizing, normalization, and tensor conversion.

feat(deployment inference): Add inference script with FFmpeg setup, multimodal preprocessing, and S3 integration for predictions.

feat(deployment): Implement deployment-ready multimodal model with frozen BERT, ResNet3D, CNN encoders and dual classification heads.

chore(deployment requirements): Add deployment requirements with pinned versions for CUDA 12.1, Whisper ASR, and inference optimizations