Deep Cody: Allow toggle in UI & implement CodyChatMemory

[abeatrix]

CLOSE: https://linear.app/sourcegraph/issue/CODY-4246/remove-deep-cody-from-the-model-list
CLOSE: https://linear.app/sourcegraph/issue/CODY-4247/add-toggle-for-35-sonnet

Deep Cody is now an experimental feature that users can toggle on/off with a feature flag enabled.
This allows us to remove Deep Cody from the model drop-down list, and makes it easier to enable Deep Cody for other models in the future.

UI Change:

• Remove Deep Cody from model list
• Stops treating Deep Cody as model
• Integrates Deep Cody toggle functionality into the chat UI
• Updates the ContextCell and HumanMessageEditor components to display and handle the Deep Cody toggle
• Adds new API methods to the WebviewToExtensionAPI interface to enable/disable Deep Cody
• Updates the ChatController to handle the Deep Cody feature flag and toggle
• Removes the automatic selection of Deep Cody as the default model
• Adds unit tests for the Deep Cody feature

Deep Cody Agent class refactor:

• Introduce CodyChatAgent as the base class for managing chat agents
• Implement CodyChatMemory to store and retrieve temporary chat context
• Refactor CodyToolProvider to use a registry-based approach for managing tools
• Add new tool - MemoryTool - a tool that allows Cody to build memory when asekd
• Improve handling of OpenContext-based tools

Add agent field to chat messages

This allows for more flexibility in enabling different chat agents.

• Added agent field to ChatMessage and SerializedChatMessage interfaces
• Updated ChatController and webview components to use the agent field
• Enabled Deep Cody context when the agent is set to 'deep-cody'
• Updated tests and other related code to reflect the new agent field

Test plan

Start branch locally and confirmed everything works as expected

Memory Tool

Changelog

[github-actions]

‼️ Hey @sourcegraph/cody-security, please review this PR carefully as it introduces the usage of an unsafe_ function or abuses PromptString.

[valerybugakov]

Tested locally, approving with some feedback inline. It works!

[valerybugakov]

To align with our recent efforts to enhance release stability and enforce stricter PR reviews, I recommend breaking PRs into smaller, focused changes where possible. This approach has several benefits: if an issue arises, we can roll back only the specific change rather than the entire function, and focused changes make the review process much easier.

Since this is an experimental feature and it's already reviewed, splitting it isn't required, but let's keep this in mind for future submissions. We can aim to keep non-generated changes in PRs to a minimum necessary for the PR's primary goal.

[abeatrix]

To align with our recent efforts to enhance release stability and enforce stricter PR reviews, I recommend breaking PRs into smaller, focused changes where possible. This approach has several benefits: if an issue arises, we can roll back only the specific change rather than the entire function, and focused changes make the review process much easier.

Since this is an experimental feature and it's already reviewed, splitting it isn't required, but let's keep this in mind for future submissions. We can aim to keep non-generated changes in PRs to a minimum necessary for the PR's primary goal.

Agree, and noted!
Thanks for the thorough review @valerybugakov 🙇‍♀️