Swap docs toolchain from MyST to Quarto

[MaxGhenis]

Part 1 of 3 (A: toolchain swap → B: graph + generator → C: docs rewrite) per PR #301's review feedback.

Scope

Swap the docs toolchain from MyST to Quarto. No content changes — all 11 existing .md pages render under Quarto unchanged.

Changes

• docs/myst.yml → docs/_quarto.yml with the same navigation, grouped into collapsible sidebar sections.
• docs/examples.md — convert the 8 {literalinclude} directives to Quarto's {.python include="..."} (the only MyST-specific construct in the current tree).
• Makefile — make docs → quarto render docs; make docs-serve → quarto preview docs.
• .github/workflows/pr_docs_changes.yaml — install Quarto via the official quarto-dev/quarto-actions/setup@v2 action.
• docs/.gitignore — track _site / _freeze / .quarto / .quarto_ipynb.

Timing

Local quarto render docs takes 4 seconds for all 11 pages on an M-series laptop (cold Quarto, no cache). CI first-run overhead from the Quarto binary install is ~30 seconds before the action's cache kicks in; overall PR-docs job stays under a minute.

Per earlier review feedback this is acceptable — within a factor of 2 of the MyST time, trading ~15 seconds of CI job time for the multi-output capability Quarto brings.

Why Quarto

• One source, two outputs (HTML site + PDF) — matters once the methodology paper lands as a chapter of the same tree.
• Computational blocks execute live Python against the installed package during build, so tutorials stay in sync with releases rather than going stale.
• Native journal templates, BibTeX citations, cross-references, Observable cells.
• The main Jupyter Book / MyST maintainer moved to Quarto; JB2's development pace has slowed.

Test plan

• quarto render docs succeeds locally (11/11 pages)
• All {literalinclude} → {.python include=} conversions produce identical rendered output
• CI passes

Follow-up PRs

• B — policyengine.graph as an optional extra (policyengine[graph]) plus the auto-reference generator under docs/_generator/
• C — docs content rewrite layered on top of Launch v4: agent-first surface, shared microsim base, dict reforms, refreshed docs #298's v4 refresh (user-journey restructure, tighter per-page length, split core-concepts into households / reforms / microsim)