Add NotebookEdit tool for Jupyter notebooks

[aakashrajput]

What

Claude Code has a dedicated NotebookEdit tool for editing .ipynb cells directly (add/edit/delete a specific cell by index, preserving notebook JSON structure, outputs, and metadata) instead of forcing the model to hand-edit raw notebook JSON through edit_file. KlaatCode has no notebook-aware tool today — editing a .ipynb currently means the model manipulating the raw JSON via edit_file/write_file, which is fragile (easy to corrupt cell structure, execution counts, output arrays).

Where to start

• New tool notebook_edit (src/tools/index.ts, following the existing tool-registration pattern): operations insert_cell / edit_cell / delete_cell, addressed by cell index, cell_type: "code" | "markdown".
• Parse/serialize with the standard nbformat JSON structure — preserve everything the tool isn't explicitly asked to change (outputs, execution_count, metadata) except when editing a cell's source clears its outputs, matching normal notebook-editing semantics.
• Route through the same sandbox/write-allowlist and permission-prompt flow as other write tools; diff preview can show the affected cell's before/after source.
• Read-side: a notebook_read or extending read_file's existing handling to summarize a notebook (cell count, types, first line of each) rather than dumping raw JSON, so the model doesn't need the whole file to decide which cell to touch.

Acceptance criteria

• Model can add/edit/delete a specific cell in a real .ipynb file without corrupting the surrounding notebook structure.
• A few unit tests round-tripping a small fixture notebook through insert/edit/delete.

Why it matters

Niche but bounded — data-science/ML users editing notebooks currently get a worse, riskier experience than plain source files.