Add a synthetic dataset for local testing

[lispandfound]

Running nzcvm generate end to end has needed the real NZCVM data root:
a DEM, a Vs30 raster, a coastline, a tomography model and a directory of
basin meshes. That makes local iteration awkward and keeps the whole
generate path out of the test suite.

nzcvm/synthetic.py defines a small analytic world over a 48 x 44 km patch
of coast: elevation falls west to east and crosses sea level three quarters
of the way across, Vs30 tracks elevation, two paraboloid basins sit inland,
and a tomography block fills the rest. Every field is closed form, so the
data regenerates identically anywhere and a test can work out the expected
answer by hand.

nzcvm synthetic serialises those fields in the formats the production
readers already accept, so the synthetic files flow through the same
converters and layers as the real ones — surfaces go out as modeller-style
HDF5 for nzcvm surface convert, the tomography block as an EP2020 CSV for
nzcvm tomography convert, and each basin as an outline plus a basement
surface for nzcvm basin main.

just synthetic chains the lot into a data root under synthetic/, which
examples/synthetic.toml then generates a velocity model from.

Testing

just synthetic                                                  # ~1 min, mostly gmsh
uv run nzcvm generate examples/synthetic.toml synthetic/model.zarr
uv run --dev pytest tests/          # 194 passed, 1 skipped
uv run --dev pytest --doctest-modules nzcvm/
uv run ruff check nzcvm/ tests/ && uv run ty check nzcvm/

tests/test_synthetic.py asserts the analytic properties the rest of the
suite leans on (the shoreline is where elevation crosses zero, a basin closes
to zero thickness on its own outline) and reads each written file back through
the matching production reader. Building the basin meshes goes through gmsh
and is left to just, which is too slow for a unit test.

🤖 Generated with Claude Code

---

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}

[github-actions]

Benchmark for 5a74360

Click to view benchmark

Test	Base	PR	%
Mesh_Point_Query/centre/4096000	245.0±1.32ns	244.2±0.62ns	-0.33%
Mesh_Point_Query/far_corner/4096000	232.2±1.00ns	231.8±1.19ns	-0.17%
Mesh_Point_Query/near_origin/4096000	280.1±3.53ns	282.3±2.10ns	+0.79%

[github-actions]

Benchmark for ca8b094

Click to view benchmark

Test	Base	PR	%
Mesh_Point_Query/centre/4096000	304.1±4.79ns	302.7±1.54ns	-0.46%
Mesh_Point_Query/far_corner/4096000	201.1±3.03ns	202.2±4.82ns	+0.55%
Mesh_Point_Query/near_origin/4096000	345.9±1.62ns	348.7±2.16ns	+0.81%