Add list_templates() to discover from_template names

[brendancol]

Summary

from_template accepts curated regions, world cities, and country codes, but the names live in private dicts (_REGIONS, _CITIES, _COUNTRY_BBOXES). Short of reading the source, there was no supported way to find out what's available. This adds list_templates() to fill that gap.

• list_templates() returns the accepted names grouped into regions, cities, and countries, each a sorted list.
• list_templates(kind) returns one group's list; a bad kind raises ValueError.
• Every name it returns is a valid from_template argument (regions/cities lowercase, countries uppercase ISO-3166 / GADM alpha-3).
• The unknown-name error and the templates reference doc now point at the helper.

Available as both xrspatial.list_templates (next to from_template) and xrspatial.templates.list_templates.

Backends

No backend dimension. list_templates returns plain lists and dicts of strings, not arrays. from_template is unchanged.

Test plan

• list_templates() keys are exactly regions/cities/countries, each equal to the sorted registry keys
• kind filter returns the right sorted list; invalid kind raises
• every listed name builds via from_template
• unknown-name error mentions list_templates
• full test_templates.py passes (63 tests), flake8 clean

Closes #3535

[brendancol]

PR Review: Add list_templates() to discover from_template names

Blockers

None.

Suggestions

None.

Nits

• xrspatial/templates.py:202 and :238-242 keep the kind names in two places: the _TEMPLATE_KINDS tuple feeds the error message, while the validation checks kind not in groups. Add a fourth group to groups and forget the tuple, and the error message goes stale. Building the message from tuple(groups) and dropping _TEMPLATE_KINDS removes the drift.

What looks good

• Every name the helper returns is a real from_template argument, and the tests check that directly: test_list_templates_grouped compares against the sorted registry keys, and test_list_templates_names_resolve builds one from each group. That's the contract that matters.
• The unknown-name error and the reference doc both point at the helper now, which is what the issue asked for.
• Leaving list_templates out of the README backend matrix is right. It returns plain strings, no backend behavior to track.

Checklist

• Algorithm matches reference: N/A (registry listing)
• Backend parity: N/A (no backend dimension)
• NaN handling: N/A
• Edge cases covered by tests: yes (invalid kind, every group resolves)
• Dask chunk boundaries: N/A
• No premature materialization: yes
• Benchmark: not needed (dict lookup)
• README feature matrix updated: yes (folded into the from_template row)
• Docstrings present and accurate: yes

[brendancol]

Follow-up review

The single nit from the first pass is resolved in 2b9c6a4: dropped the _TEMPLATE_KINDS constant and the error message now reads its options from tuple(groups), so the message can't drift from the actual groups.

No new issues. Tests still pass (9 list_templates/error tests green, flake8 clean). Nothing else outstanding.