[Feature Request]: video Embeddings : chromaDB also for videos

### Describe the problem

For up until now chromaDB is only for text, from this feature it will be the solution for videos as well

like "Chat With Video" / "Talk To Video"

### Describe the proposed solution

**function that creates embeddings for videos**

### steps :

video ==> audio ==> text ==> nltk(sent_tokenize) ==> vector (steps followed in text ) -- (creating collection for sentences in video )

### sample code :

from moviepy import VideoFileClip
import speech_recognition as sr

def extract_audio_from_video(video_file_path , audio_file_path):
'''

desc : extracts audio from video_file_path and stores it in audio_file_path

input :
video_file_path : path to video file
audio_file_path : path to audio file

output :

'''

video = VideoFileClip(video_file_path)
video.audio.write_audiofile(audio_file_path)

def transcribe_audio_to_text(audio_file_path , text_file_path):
'''
desc : transcribes the audio to text

input :
audio_file_path : path to audio file
text_file_path : path to text file to save the text

output :
'''
recognizer = sr.Recognizer()
with sr.AudioFile(audio_file_path) as source:
audio_data = recognizer.record(source)
text = recognizer.recognize_google(audio_data)
with open(text_file_path , "w") as file:
file.writelines(text)
return text

video_file_path = "samples.mp4"
audio_file_path = "temp_audio.wav"
text_file_path = "text_of_video.txt"

extract_audio_from_video(video_file_path, audio_file_path)
transcription = transcribe_audio_to_text(audio_file_path , text_file_path)
print("Transcription:")
print(transcription)

and I would like to contribute to this feature

### Alternatives considered

_No response_

### Importance

nice to have

### Additional Information

I would like to work on it!

No repository files, tests, or entry points are identified. Start by reviewing the existing text embedding and Chroma collection flow, then determine how the proposed video-to-audio-to-text pipeline would fit it. The issue needs an agreed design, dependencies, and acceptance criteria before completion can be measured.

Written by the indexing model from the issue text.