Require regionmask >=0.13

[hdrake]

What this changes

regionate/grid_conform.py calls regionmask.mask_geopandas (three sites), which
regionmask did not gain until 0.6. The dependency was declared with no lower bound
in both pyproject.toml and conda/meta.yaml, so any resolver was free to pick a
regionmask that lacks the function.

That is not hypothetical. Building the conda-forge feedstock for 0.5.6 solved to
regionmask 0.4.0 (2019), which contains no mask_geopandas — its __init__
still exports the Regions_cls / create_mask_contains API. The resulting package
imports fine and then dies at mask time. The feedstock build happened to fail for an
unrelated reason (pip check rejecting that package's stale Tag: cp36-none-any
wheel metadata), which is the only reason it was not published.

Changes

• pyproject.toml: regionmask → regionmask >= 0.13
• conda/meta.yaml: regionmask → regionmask >=0.13

Floor set at 0.13 rather than the bare API minimum of 0.6 because 0.13 is the version
the suite is actually run against. No upper bound — a ceiling here would recreate the
kind of stale cap that forces resolvers backwards.

Validation

• pytest passes (2 passed) in a clean environment resolving to regionmask 0.13.0,
  geopandas 1.1.4, rasterio 1.5.0, numpy 2.4.6, sectionate 0.3.4.
• A cross-platform conda create --dry-run --platform linux-64 with regionmask>=0.13
  solves cleanly: regionmask 0.13.0, rasterio 1.4.4, pooch 1.9.0, numpy 2.4.6.
• No test is added: the change is a dependency floor, and CI installing the package
  and running the suite is what exercises it. A test asserting a version number would
  only restate the metadata.

Follow-ups not in this PR

conda/meta.yaml is stale in two other ways, left alone to keep this diff single-purpose:
{% set version = "0.5.5" %} with a 0.5.5 sha256, and sectionate >=0.3.3 where
upstream now says >=0.3.4,<0.4.

[hdrake]

Looks good.