Dedicated session for Assistant

Context

Related to https://github.com/posit-dev/positron/issues/14771

Currently Assistant shares a single console with the user. This makes it difficult to review and audit work written by hand vs. agent.

In an Agent/ Assistant first chat mode, this becomes more important for spot checking AI generated work so a user can decide what to merge into their work

AC

1. When Assistant executes code, the user's console does not get flooded with the Assistant's code execution
2. If the user wants to review a log of the Assistant's code execution, they can easily access it

Potential implementation

1. Dedicated agent session - Agent gets its own console/kernel instance; user's interactive console stays untouched unless they explicitly pull agent output in.

• we'll need to reconcile environment changes i.e. package installs

2. Shared kernel, separate execution log - Same underlying kernel/environment but the console view/history is partitioned by actor

• we'll need to tag each console entry as user/assistant

Open questions

1. Are there startup and performance concerns with spinning up separate sessions
2. Is this attractive for some of the customer asks around constraints or sandboxing of the Assistant R/Python session? Can we launch a session in a sandbox bubblewrap style or Landlock?