Speed up whisper: alignment model

[oserikov]

dream/annotators/vidchapters_service/server.py

Line 103 in c779fb9

align_model, metadata = whisperx.load_align_model(language_code=asr['language'], device = DEVICE, model_dir=os.path.join('/src/aux_files', MODEL_DIR))

Let's move this out of the function