fix: protect Codex vterm marker extraction from scrollback loss

Raise the vterm scrollback floor for Codex buffers before libvterm creation,
including dir-local overrides, and report unsafe settings in doctor.

fix: add mode line help to codex top mode

Expose the available top-mode key bindings directly in the buffer mode line
and bump the package version.

feat: add dashboard to monitor active Codex sessions

Introduce my-codex-top , a tabulated list dashboard showing session
attributes (PID, branch, git status, prompt count, and activity age). Add
keybindings to visit sessions, kill processes, view diffs, and refresh the
list.

Close #1

feat(agent): support Google Antigravity and multi-agent profiles

Introduce agent profiles to allow running Google Antigravity alongside
OpenAI Codex. Users can choose the default agent or select one per
session. The active agent is tracked per project, and session buffer names
are prefixed accordingly. Also add tests, documentation, and doctor checks
for the new configurations.

fix: switch selected sessions in the Codex window

Route session-list selections to the active Codex session window and keep
edit-window layout metadata in sync when switching sessions.

feat: list open Codex sessions

Add a session listing command under F8 S l that shows live session buffers
with their names, access modes and project roots. Exclude stale renamed
buffers and document the new binding.

feat: add local Git diff view commands

Add persistent diff-mode buffers for worktree and staged changes, exposed
via F8 v and F8 V alongside the existing Codex review and Ediff Git helpers.

fix: avoid prebinding vterm-shell

Read vterm-shell dynamically so vterm can keep its own shell fallback. Add
tests covering shell command generation and the unbound load case.

feat: show token estimates in prompt previews

Add an initial prompt-size label to the preview header using the existing
approximate token counter, so users can judge context size before sending.

Document the preview estimate, bump the package version, and cover the
header text with an ERT test.

feat: guard against oversized Codex prompts

Add configurable warning and hard-limit thresholds before sending prompts to
Codex, including an approximate token count in the confirmation message.

Document the new prompt-size options and cover the guard behaviour with ERT
tests.