Render GPlates coastlines as filled spherical surfaces

[sghelichkhani]

Summary

Reworks CoastlineVTKFile so GPlates coastlines render as proper triangulated surfaces on the sphere. The previous pipeline wrote each coastline as a single polygon cell, which VTK fan-triangulated in a best-fit plane — producing crumpled continents with empty interiors and spurious triangles bridging date-line-split pieces.

The new pipeline:

• Splits polygons that cross the antimeridian with DateLineWrapper and triangulates each sub-polygon independently, so continents no longer bridge across disjoint pieces.
• Triangulates in (lon, lat) with mapbox_earcut rather than in VTK's best-fit plane — no more fan wedges for concave or large continents.
• Adaptively refines the mesh on the sphere with subdivide_adaptive, re-projecting new vertices onto the sphere after each pass, so long thin earcut ears become uniform patches that hug a great-circle arc.
• Re-orients every triangle normal outward regardless of the source polygon's winding, so ParaView renders them from outside the globe.

Demo and tests

Adds examples/coastlines_vtp/ with a jupytext demo, four regression tests, and a baseline expected.pkl. The test data (Muller et al. 2019 rotations + coastlines, ~4 MB) is hosted on the gadopt DO Spaces CDN at g-drift/test_data/coastlines_vtp/ and auto-downloads into gdrift/data-gplates/ on first test run. The demo uses the same three-step resolver (env var → local gtrack/ copy → CDN) so it works on a fresh clone without any setup.

Packaging

Adds a gplates extras group (pyvista, mapbox_earcut) to pyproject.toml. pygplates is noted in a comment rather than listed, since it is distributed by the GPlates team rather than via PyPI.

Known limitation

Polygons that enclose a geographic pole (notably Antarctica) are not rendered correctly: earcut in (lon, lat) does not close the ring across the pole, so Antarctica appears as a strip with a polar hole. Documented in the module docstring; fixing it would require injecting the pole as an extra vertex or meshing in a polar projection and is out of scope.

Test plan

• make lint clean on changed files
• pytest examples/coastlines_vtp/test_demo.py — 4/4 pass locally against auto-downloaded CDN data
• Demo executes end-to-end and produces a .pvd + 6 .vtp files that render correctly in ParaView
• CI on this branch (auto-download + skip-without-deps path)