Ollama: add image upload support to multimodal models

[abeatrix]

NOTE: This is a proof of concept on supporting multimodal in Cody. The UI will need to be updated to make it more obvious to users when an image has been uploaded for the current chat.

• Add support for uploading images to the Ollama chat client
• Update the CompletionParameters interface to include an images field
• Modify the SimpleChatModel to handle image uploads and provide the base64-encoded images to the chat client
• Update the SimpleChatPanelProvider to handle the new 'chat/upload-image' command from the webview
• Add an image upload button to the Toolbar component when using an Ollama model that supports multimodal inputs

Multimodal models from Ollama that we currently support: llava & bakllava

Demo

Loom: https://www.loom.com/share/4d2b72de1385436a960b0775434b933e

Follow-up works/ideas

• UI feedback when the user has uploaded an image
• Allow users to remove an uploaded image in the UI
  • Currently, the image will only be removed after they have submitted the question
• Look into supporting other media next?

Test plan

1. Download Ollama
2. Download the supported Ollama models: llava or bakllava
3. Confirm your Ollama is running
4. Build and start Cody from this branch
5. Verify you can find the model you just downloaded in the model dropdown menu
6. Choose the supported model
7. Verify a paperclip icon is showing up when you selected the supported model
8. Click on the paperclip icon to select an image (png, jpg, SVG files only)
9. Ask Cody to explain the image you attached
10. Verify the response is correct

Example

Left image was shared with Cody, right image was built with the code provided by Cody:

[PriNova]

@abeatrix This is amazing even if the open-source vision models are benching a lower than GPT-4o or Sonnet 3.5. If, in the future, the SG endpoint will be compatible later with Sonnet 3.5 and other multi-modal models, this tops that all.