[v0.5] Add terminal pane as a window leaf

[owainlewis]

Goal: Add an embedded terminal pane that fits the existing window model instead of creating a separate AI subsystem.

Context: The PRD’s agent strategy is composition: a coding agent runs in a real terminal pane and edits files on disk.
This should land only after the split/window model exists, because the pane should be another window leaf drawn into a rectangle.

Proposed approach:

• Add a terminal content type beside file-buffer windows.
• Spawn a shell in a PTY with the repo working directory as the default directory.
• Route input to the PTY when the terminal pane is focused.
• Render the PTY grid through the same renderer boundary used for editor windows.
• Resize the PTY when its pane rectangle changes.
• Add one command to create/open a terminal pane from the current layout.
• Keep the first pass shell-only; running Claude Code or another agent is just what the user types inside the pane.

Acceptance criteria:

• A command creates a terminal pane in the current window layout.
• Focused terminal panes receive typed input and display shell output.
• Focus can move between file windows and terminal panes.
• Resizing splits updates the terminal pane size without corrupting the display.
• Closing the pane cleans up the child process or leaves a clear, intentional process policy.
• Exiting Cortex restores the outer shell cleanly.
• Existing file-window editing, saving, splits, tabs, and reload behavior still work.

Verification:

• Run cargo test.
• Manually open a terminal pane, run pwd, run a command that prints multiple lines, resize the terminal, focus a file pane, edit/save, and quit.
• Manually confirm the outer shell is usable after exit.

Out of scope:

• AI provider integration.
• Agent protocol APIs.
• Chat UI.
• Multiplexer features beyond this editor pane.
• Scrollback search.
• Multiple shell profiles.