Write unit tests CI, and pre-commit for main functionality

[jcsmithhere]

⚠️ Preliminary plan — review before executing. The task list below was
drafted automatically to scope this issue. It must be reviewed and approved
before any of it is implemented; treat it as a starting point, not a
committed plan.

Plan: unit tests, CI, and pre-commit

The package currently has no test suite, CI, or linting, so this covers all three. A guiding constraint: most data loaders hit live network APIs (neo-bolide, JPL CNEOS, GMN, IAU MDC), so unit tests must mock the network or use small packaged CSV/pickle fixtures rather than making real requests in CI.

1. Test scaffolding

• Add pytest (+ pytest-cov) as a test optional-dependency group in pyproject.toml.
• Create a tests/ package with a conftest.py providing shared fixtures (e.g. a small BolideDataFrame loaded from a committed CSV fixture).
• Add small offline data fixtures under tests/data/ (a trimmed GLM CSV, a USG sample, a GMN sample) so loaders can be tested without network access.

2. Core BolideDataFrame tests (bdf.py)

• Construction from each offline source: csv, pickle (and mocked glm/usg/gmn/remote).
• Subclass preservation: indexing/filtering returns a BolideDataFrame, not a bare GeoDataFrame (_constructor, __getitem__, force_bdf_class).
• annotate() adds the expected columns (phase, moon_fullness, solarhour, sun_alt_obs/_app).
• filter_date (start only, end only, both; tz-naive vs tz-aware input; inplace True/False).
• get_closest_by_time, get_closest_by_loc (haversine correctness), and get_closest dispatch logic.
• filter_boundary and filter_observation (incl. intersection vs union; GOES-16/17 selection; the obs.csv header=1 parsing).
• filter_shower (with a fixture ShowerDataFrame to avoid IAU network calls).
• augment cross-matching (time/score thresholds; intersection/outer/left).

3. Supporting-module tests

• fov_utils.get_boundary: valid names return polygons, invalid raise ValueError, intersection/union/collection options, CRS conversion.
• astro_utils: haversine against known distances, _distance_metric, solar/lunar helpers, sol_lon_to_jd/sol_lon_to_datetime.
• sources: each loader's parsing logic with mocked HTTP responses (e.g. responses/requests-mock or monkeypatched download).
• sdf.ShowerDataFrame.get_dates with a fixture.
• utils helpers (reconcile_input, make_points, str_to_list).

4. CI (GitHub Actions)

• Add .github/workflows/tests.yml running the suite on push and PR across a Python version matrix (respect requires-python >=3.7).
• Install system libs needed by cartopy/shapely (PROJ, GEOS) in the workflow, or use a conda-based setup.
• Upload coverage (e.g. Codecov) and add a status badge to the README.
• Mark the suite as a required check so it must pass before merging to master.

5. Pre-commit

• Add .pre-commit-config.yaml with a linter/formatter consistent with existing style (~100-char lines) — e.g. ruff/flake8 plus whitespace/EOF hooks.
• Add a hook (or .gitignore enforcement) to keep editor cruft (*.swp, *~) out of commits.
• Document pre-commit install and how to run the tests in CLAUDE.md and the docs once in place.