Fix shared-CONSTANTS state leaks in doctests, wire up doctest CI

[derek73]

Summary

• Several doc/docstring examples mutated the shared CONSTANTS singleton (capitalize_name, force_mixed_case_capitalization, empty_attribute_default, titles.clear()) without resetting it afterward. Since Sphinx's doctest groups only isolate local variables (not global state), these mutations leaked forward across the entire doc build in an order-dependent way, causing several previously-undetected doctest failures:
  • JEFFREY (JD) BRICKEN example wrongly showing capitalized output
  • as_dict() returning None instead of '' for empty fields
  • "Sir Bob Andrew Dole".initials() returning 'S. B. A. D.' instead of 'B. A. D.' (because CONSTANTS.titles had been cleared and never restored, so "Sir" leaked into first_list)
• SetManager.__repr__ printed a plain Python set, whose iteration order depends on per-process string hash randomization — the same code produced different-looking output on every run, making the {'right', ..., 'tax'} ellipsis-style doc examples fundamentally unreproducible.
• Two smaller doc-only bugs in customize.rst: an emoji example showed repr(hn) output but was written as if it were str(hn); CONSTANTS.titles.clear() was documented as producing no output, but clear() returns self (chainable) and so prints SetManager(set()).

Changes

• nameparser/config/__init__.py: reset CONSTANTS at the end of each docstring demo; SetManager.__repr__ now sorts elements so its output is deterministic across processes.
• docs/usage.rst: reset CONSTANTS.capitalize_name / force_mixed_case_capitalization after their respective demo blocks.
• docs/customize.rst: switched the "Hon" removal / clear-titles demo to a local Constants() instance instead of the shared singleton, since that section isn't actually teaching shared-singleton behavior (the "Module-level Shared Configuration Instance" section further down still correctly uses the shared singleton, since demonstrating that gotcha is its whole point); fixed the emoji example to call str(hn); updated the SetManager ellipsis examples to match the now-sorted repr.
• .github/workflows/python-package.yml: added sphinx-build -b doctest docs dist/doctest and python -m doctest README.rst so this class of regression is caught automatically going forward, across all 5 Python versions in the CI matrix.

Test plan

• pytest -q — 1270 passed, 4 skipped, 22 xfailed (unchanged baseline)
• sphinx-build -b doctest docs ... — 186/186 passed (previously multiple failures across usage.rst, customize.rst, and parser.py's own docstrings)
• python -m doctest README.rst — 10/10 passed
• Verified SetManager.__repr__ produces identical output across separate process invocations (previously varied per run)

🤖 Generated with Claude Code