anwer_clip_ollama: new operator to do vqa on clips

[geoffroy-noel-ddh]

This is similar to BVQA and the answer_transcription_ollama.

This assumes that ollama support video input and has compatible VLMs. E.g. qwen3-vl

One hurdle is finding a method to efficiently pass a video to ollama using FrameSense container architecture.

[geoffroy-noel-ddh]

Ollama doesn't support video as an input, only images.

This operator then becomes answer_clip_qwen3vl or something. Would be nice to make it work on three units: video, clips and frames.