…ading

Switch open_notebook_window, spawn_new_notebook, and complete_onboarding
to use create_notebook_window_for_daemon with OpenMode::Open/Create.

- open_notebook_window: no longer calls load_notebook_state_for_path or
  parses .ipynb locally. Daemon loads the file.
- spawn_new_notebook: no longer calls NotebookState::new_empty_with_runtime.
  Daemon creates the empty notebook.
- complete_onboarding: same as spawn_new_notebook but with working_dir
  from ensure_notebooks_directory().

The old create_notebook_window/create_notebook_window_with_label are
still used by run() startup and session restore (converted next).

…on-owned loading

The main window and additional session windows now use daemon-owned
loading instead of local NotebookState construction:

- CLI path → OpenMode::Open (daemon loads .ipynb)
- Session with path → OpenMode::Open (daemon loads)
- Session untitled with env_id → OpenMode::Restore (reconnect to
  existing daemon room via old handshake)
- No session, no path → OpenMode::Create (daemon creates empty notebook)

Additional session windows use create_notebook_window_for_daemon
instead of create_notebook_window_with_label, avoiding
load_window_session_state and local .ipynb parsing.

The async daemon sync init block now dispatches to the appropriate
init function based on OpenMode instead of extracting cells/metadata
from NotebookState.

create_window_context (old version taking NotebookState) is now unused.

…oading

When a .ipynb file is opened via Finder/dock icon on an empty main window,
the handler now updates the context path and spawns
initialize_notebook_sync_open instead of calling load_notebook_state_for_path.

No local .ipynb parsing for the reuse-main-window case. The new-window
case already goes through open_notebook_window (converted earlier).

Replace the get_cells() → FrontendCell conversion → initialize_notebook_sync
chain with initialize_notebook_sync_open. The daemon just wrote the file,
so OpenNotebook loads it right back without carrying cells across.

This eliminates:
- The last get_cells() call site on NotebookSyncHandle
- The last FrontendCell usage in lib.rs
- The CellSnapshot → FrontendCell conversion code
- notebook_state.path sync-back (no longer needed)
- Local notebook_id derivation in save-as (daemon canonicalizes)

Dead code warnings now show: derive_notebook_id, notebook_state_for_window,
create_notebook_window, create_notebook_window_with_label, create_window_context
are all unused.

For saved notebooks, use initialize_notebook_sync_open (daemon reloads
from disk). For untitled notebooks, use the old handshake with the
notebook_id (env_id) so the daemon can find its persisted Automerge doc.

save_session now reads path from context.path, notebook_id from
context.notebook_id (env_id for untitled), and runtime from
context.runtime. No more locking NotebookState.

Remove load_window_session_state — session restore now dispatches
directly to OpenMode in run() without constructing NotebookState.

Remove all code that is no longer called now that every entry point
uses daemon-owned loading:

Deleted functions:
- derive_notebook_id (daemon canonicalizes paths)
- notebook_state_for_window (no code reads NotebookState)
- create_notebook_window / create_notebook_window_with_label (replaced by _for_daemon)
- create_new_notebook_state (daemon detects project files)
- load_notebook_state_for_path (daemon loads .ipynb)
- create_window_context (replaced by _for_daemon)

Removed from WindowNotebookContext:
- notebook_state field (session save reads Arc fields)

Simplified initialize_notebook_sync:
- Removed initial_cells and initial_metadata params (all callers passed empty)
- Removed cell population loop (daemon populates the room)
- Removed FrontendCell import (no longer used in lib.rs)

The NotebookState stub in create_window_context_for_daemon is also
gone — the context no longer holds any NotebookState.

Net: -268 lines. Zero warnings.

…_window_for_daemon

The placeholder_id was computed from path alone, which is None for
Restore and Create variants — resulting in an empty notebook_id in
the context. When reconnect_to_daemon later reads context.notebook_id,
it connected to a room with empty ID instead of the env_id.

Now derives placeholder_id from the OpenMode variant:
- Open: canonical path (same as before)
- Restore: notebook_id from the variant (the env_id)
- Create: empty (daemon generates UUID)

This fixes session restore for untitled notebooks showing as empty
on reconnect.

new_fresh previously deleted all persisted docs unconditionally,
treating .ipynb as the sole source of truth. For untitled notebooks
(UUID-based notebook_ids), the persisted Automerge doc is their only
content record — there's no .ipynb on disk.

Now: for untitled notebooks, load the persisted doc if it exists so
content survives daemon restarts. For saved notebooks, still delete
the stale persisted doc and load from .ipynb (unchanged behavior).

Add test verifying persisted docs are loaded for UUID notebook_ids.