fix: resolve ruff and mypy from the generator's own environment

[goduni]

Both tools are invoked at runtime, but they were looked up with shutil.which(), which searches PATH. That is the wrong place: uv tool install and pipx expose only the console scripts a distribution declares, so the copies installed alongside the generator are not on PATH and an unrelated system install wins instead. The bug reproduced in this very repo, where PATH resolved to ruff 0.15.10 / mypy 1.20.0 while the pinned environment held 0.15.15 / 2.1.0 -- generated code was being formatted and checked by versions nobody selected.

Add tooling.py, which searches the environment the generator was installed into before anything ambient. Its first candidate is derived from this module's own location, because ruff arrived through the same install; sysconfig's default scheme cannot stand in for it, since under pip install --user that names the system bin while the generator and its ruff live under the user scheme, letting a too-old system ruff win. The remaining candidates are the active environment, the interpreter's own directory, the user scheme and the base prefix, with PATH last.

ruff is resolved to a binary path rather than run as python -m ruff, which measured 4.4x slower per call (12.7ms -> 56.5ms) and is invoked twice per generated file. mypy runs as python -m mypy, a single call where the interpreter also determines what it resolves imports against.

Promote mypy from the check extra to a core dependency, since --check is documented as a headline feature and the extra left it absent by default. The extra stays, now empty, so existing [check] installs keep resolving.

--check gains --python-executable pointing at an activated virtualenv when that differs from the generator's own environment. Without it, moving mypy off PATH would have regressed standalone installs: the generator's environment has ruff and mypy but not unihttp or the serializer, so every generated import would fail to resolve. Environment roots are compared rather than interpreters, because venvs built from one base python share a bin/python symlink target and would otherwise look identical.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (9ca381d) to head (757b137).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main        #6   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           29        30    +1     
  Lines         2557      2604   +47     
=========================================
+ Hits          2557      2604   +47     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.