[FEATURE] Agent teams

## Feature Description

### Support for agent teams

As far as I can tell from the docs and the code, there would be issues when using agent teams with interacting with the worktrees and the neovim instance.

## Use Case

Parallel worktrees with agents assigned to them via agent teams for larger projects

## Proposed Solution

We could have a separate tab for each parallel worktree, where the tab's buffers are scoped to that worktree.

Main issue here is, how do we organize our directory structure so that when you are navigating between files, it is easy to see what is in what worktree? The only way I can think to do this is if you spawn your nvim instance from the parent directory. For example, if I have a `project` directory for my main work (without worktrees), I would need a `project_wrapper` directory (terrible name, but you get the idea) that has the `project` dir and the worktree dirs that the agents are spinning up as siblings.

The tab-per-worktree UI is one problem, but the bigger issue is probably the WebSocket server itself. Right now the plugin runs one server with one lock file. Even if we get tabs working perfectly, the worker agents spawned by agent teams would all be hitting the same server, and the diff/context routing has no concept of "this diff belongs to this worktree's tab." That's a pretty fundamental architectural thing to solve.

Also worth noting — the lock file convention (`~/.claude/ide/[port].lock`) is Anthropic's, not this plugin's. If agent team workers don't know how to look for separate lock files per worktree or per session, there's only so much the plugin can do here. Might be worth checking if this is actually something Anthropic needs to address on the CLI side first (maybe the lock file protocol needs a worktree discriminator or something).

One possible incremental step: support spinning up multiple WebSocket servers (one per worktree), each with its own lock file, under `provider = "none"`. That way people using the tmux approach (like me) could at least get IDE integration per-pane without needing the full tab-scoping solution. Less ambitious, but probably more realistic as a first pass.

Has anyone looked at how the VS Code extension handles agent teams? That would tell us a lot about whether the protocol already supports this or if it's a gap everywhere.

EDIT: Turns out Anthropic doesn't support agent teams in the VS Code extension! So that's telling...

## Alternatives Considered

What I do now is just...not use agent teams. I have to coordinate in separate tmux windows a separate nvim + agent.

No file, test, or entry point is named. Start by tracing the WebSocket server, lock-file handling, and provider="none" path; a possible first milestone is multiple per-worktree servers with separate lock files, while the broader tab-scoping and protocol questions remain unresolved.

Scritto dal modello di indicizzazione a partire dal testo della issue.