chore(ci): delegate notebook execution and caching to shared mkdocs-deploy composite action

[MAfarrag]

Context

Each docs CI job in .github/workflows/github-pages-mkdocs.yml previously carried a 4-step
preamble (setup-python → restore .jupyter_cache → pip install jupyter nbconvert ipykernel
→ run scripts/prep_notebooks.py) before handing off to the shared mkdocs-deploy composite
action. That logic is now available directly inside the composite via a notebooks-path
input, so every serapeum-org repo using the composite gets the same cache semantics
without maintaining a local copy.

Tracks the work in PR #184 on branch ci/move-notebook-ci-to-github-actions.

Problem / Current Behaviour

• Three jobs (deploy-pr, deploy-main, deploy-release) each duplicate ~15 lines of
  notebook-prep setup in the workflow file.
• scripts/prep_notebooks.py (90 lines) is a repo-local re-implementation of the
  cache-aware notebook prep logic that now lives in the shared composite action.
• Drift risk: any future improvement to the prep logic (e.g. cache-key strategy, exec
  timeouts) must be applied in every repo.

Affected locations

File	Symbol	Notes
.github/workflows/github-pages-mkdocs.yml	deploy-pr, deploy-main, deploy-release	Remove setup-python / cache / prep steps; pass notebooks-path to composite.
scripts/prep_notebooks.py	module	Delete; responsibility moves to composite action.
mkdocs.yml	mkdocs-jupyter plugin block	Remove now-stale comment referencing prep_notebooks.py.
pyproject.toml	[project.optional-dependencies].docs	Add nbconvert, ipykernel so local mkdocs serve still renders outputs.
uv.lock	—	Refresh lockfile with the two new docs deps.

Motivation Example

Before (per job, ~15 lines):

- uses: actions/setup-python@v5
  with:
    python-version: '3.12'

- name: Restore jupyter notebook cache
  uses: actions/cache@v4
  with:
    path: .jupyter_cache
    key: jupyter-cache-${{ hashFiles('docs/notebook/**/*.ipynb', 'src/statista/**/*.py', 'pyproject.toml') }}
    restore-keys: |
      jupyter-cache-

- name: Prepare notebook outputs (execute uncached, merge cached)
  run: |
    pip install -e . jupyter nbconvert ipykernel
    python scripts/prep_notebooks.py

- uses: serapeum-org/github-actions/actions/mkdocs-deploy@mkdocs/v1
  with:
    trigger: 'pull_request'
    ...

After (per job):

- uses: serapeum-org/github-actions/actions/mkdocs-deploy@mkdocs/v1
  with:
    trigger: 'pull_request'
    notebooks-path: 'docs/notebook'
    ...

Proposed Solution

1. In .github/workflows/github-pages-mkdocs.yml, drop the setup-python, actions/cache,
   and prep_notebooks.py steps from all three jobs.
2. Add notebooks-path: 'docs/notebook' to the serapeum-org/github-actions/actions/mkdocs-deploy@mkdocs/v1
   invocation in each job.
3. Delete scripts/prep_notebooks.py.
4. Remove the now-stale mkdocs-jupyter plugin comment referencing the deleted script in
   mkdocs.yml.
5. Add nbconvert and ipykernel to [project.optional-dependencies].docs in
   pyproject.toml so uv run --active mkdocs serve continues to render notebook outputs
   locally.
6. Refresh uv.lock with uv lock.

Out of Scope

• Upstream changes to serapeum-org/github-actions/actions/mkdocs-deploy itself — those
  live in the composite action repo.
• Any documentation-content changes.
• Migrating other serapeum-org repos to the same pattern.

Effort Estimate

Size: S
Rationale: Half-day task, mostly deletion; implementation already sits in PR #184.

Definition of Done

• deploy-pr, deploy-main, and deploy-release jobs all succeed on PR move the ci running of the notebooks and caching them to github-actio… #184
• First run of deploy-pr executes notebooks; a re-run on the same SHA is served
  from the composite's cache (no re-execution)
• scripts/prep_notebooks.py is removed from the tree
• uv run --active mkdocs serve still renders notebook outputs locally
• uv.lock revision field change (3 → 2) is either confirmed intentional or
  restored to 3
• PR move the ci running of the notebooks and caching them to github-actio… #184 merged to main