feat(client): add video_url content part for multimodal video input

[octo-patch]

Reason: Add a video content-part type and request serialization so the OpenAI-compatible chat schema can express video input, preserving existing text/image handling.

Changes:

• Add ContentPartVideoURL content-part type and a VideoURL payload struct (mirroring ImageURL) in internal/client/types.go, serialized as the OpenAI-compatible video_url field.
• Wire video attachments into the orchestrator's Submit path alongside images: files detected as video/* are encoded as video_url content parts when the model supports vision; otherwise they fall back to a text note (parallel to the existing image behavior).
• Allow the TUI file picker to attach video files using the same vision-capability gate as images, and keep media attachments consistent on model changes.

Checks:

• go build ./...
• go vet ./internal/client/ ./internal/orchestrator/ ./internal/tui/
• go test ./internal/client/ ./internal/orchestrator/ ./internal/tui/

[mlhher]

Thank you for your contribution. I see that you removed the PR template.
As thus I must ask you to manually confirm the CLA.

Please read the CLA here: Contributor License Agreement
If you agree, please reply to this thread with the following exact phrase:
I have read and agree to the terms of the Contributor License Agreement.

Will be tested and if nothing comes up can be merged afterwards. Thank you!

[giveen]

I'm confused on what an IDE needs with video input.

[mlhher]

There are some code -> video workflows where this can come in handy, also generally I think it can be helpful to attach videos as workflows instead of manually explaining things. It is a minimal targeted change and should equally support Gemma, I am not against it.

[giveen]

Okay cool, I was just really curious on how this would work