Update design patterns and requirements to be consistent with Network Wrangler

[e-lo]

What we should update:

• CI/CD pipeline using GH Actions
• Python 3.10+ style type hints
• pathlib not os
• Installation docs
• robust versioning
• data class-based parameters
• pandas 3+ safe
• modern geopandas
• docs that document the API and use cases
• typing for output using pandera
• testing using same fixtures
• add benchmark tests

Keep the same:

• external API

[e-lo]

All items in the checklist have been addressed in PR #11. Here's a status summary:

✅ Completed:

• CI/CD pipeline — 5 workflows (push, pullrequest, prepare-release, publish, clean-docs) + 5 reusable composite actions; Python 3.10–3.13 matrix; benchmark comparison and coverage comments on PRs; OIDC trusted publishing to PyPI/TestPyPI

• Python 3.10+ type hints — from __future__ import annotations added to all source modules; removed typing.Optional/Dict/List/Union in favour of X | Y syntax across project.py, roadway.py, transit.py

• pathlib not os — replaced all os.path.join/splitext/basename with Path operations in project.py, roadway.py, transit.py; import os removed from all three

• Installation docs — README.md and docs/ fully updated to uv-first install; conda references removed

• Robust versioning — static version + setuptools_scm in build system; prepare-release.yml validates __version__ matches tag and checks PyPI for duplicates before publishing

• Dataclass-based parameters — Parameters rewritten as pydantic dataclass hierarchy (TimePeriodsConfig, CategoriesConfig, Parameters) matching network_wrangler's pattern

• Pandas 3+ safe — all inplace=True calls replaced with assignment-style operations (8 in roadway.py, 4 in project.py)

• Modern geopandas — no deprecated patterns; geopandas>=1.1.1 required

• Docs — mkdocs with mkdocstrings; api.md auto-generates from all modules; new development.md covers setup, testing, benchmarks, and release process

• Typing for output using pandera — CubeLinksTable and CubeNodesTable DataFrameModel schemas are the single source of truth for column types; coerce_df_to_model() utility replaces all manual astype() loops; old convert_int/convert_bool/fill_na functions deprecated

• Testing using same fixtures — tests/conftest.py with session-scoped fixtures mirrors network_wrangler pattern

• Benchmark tests — tests/test_benchmark.py with 3 benchmark classes; standalone benchmarks/ scripts

• External API preserved — Parameters, Project, StandardTransit, get_shared_streets_intersection_hash all still exported from cube_wrangler