Docs: uv-first install guide, chain-first tutorials, and fixes

[peterrrock2]

Summary

Major rewrite of the installation guide to give better instruction on how to work with virtual
environments. Changes to the other tutorials were intentionally kept minimal; the MarkovChain
examples were converted to the incremental "build the chain first" style, but rewording of the
surrounding narrative was limited to what was necessary.

Docs are ready for more serious edits from @chief-dweeb after this. It was faster for me to just
make the minimal changes that I wanted than it was for me to add the notes in.

Brings the user guide in line with the 1.0.0 API and tooling: an installation guide rewritten around
uv, the tutorials converted to the incremental "build the chain first" style, and a set of small doc
and repository fixes. Documentation-only. Stacks on misc/odds-and-ends-pt2.

Why

Three gaps had opened up between the docs and the current library:

• The install guide still assumed a hand-managed python -m venv and contradicted itself once uv
  guidance was added.
• The tutorials still constructed MarkovChain all at once, rather than the incremental paradigm
  introduced in recom.ipynb.
• The README advertised a gerrychain[geo] extra that no longer exists, and a stalled ReCom chain
  had no user-facing explanation.

Changes

• Rewrote docs/user/install.md around uv (uv init / uv add, uv run, git installs), keeping
  pip + python -m venv as the alternative, and added a "Checking Your Installation" section with
  copyable preflight checks and a notebook-kernel check.
• Converted the MarkovChain examples in quickstart.ipynb, data.ipynb, geometries.ipynb,
  reproducibility.md, and generate_recom_assets.py to instantiate the chain first and configure
  it incrementally, with the surrounding narrative rewritten to explain each piece as it is
  attached.
• Added a "When epsilon is the real problem" section to recom.ipynb explaining that a stalled
  chain means cuttable trees are rare, not impossible.
• Removed the obsolete gerrychain[geo] and geo_settings.txt instructions from the README, and
  gitignored the local PA_VTDs.json tutorial copy and the PA_output.jsonl that data.ipynb
  writes when executed.
• Added the python notebooks into the formatting task in the makefile.

Testing

• Docs build cleanly (make docs, warnings-as-errors)
• Manually tested relevant behavior

Ran make docs and make docs-test, confirmed committed notebooks carry no outputs, re-executed
each converted notebook into the ignored MyST-NB cache, and ran the install commands and preflight
checks end-to-end in fresh projects across the released, git, and development versions.