fix(cql2-json): accept a property reference as an interval bound

[C1-BA-B1-F3]

Fixes #116.

An interval bound was assumed to be a date/time string, so

{"interval": [{"property": "start"}, {"property": "end"}]}

died in walk_cql_json with AttributeError: 'dict' object has no attribute 'split'. That shape is CQL2 conformance test 41 and is how T_CONTAINS(interval(a, b), interval(t1, t2)) is written (the linked pycql2 schema in the issue thread allows it too).

The fix walks a non-string bound like any other expression, so it becomes an ast.Node that the evaluator resolves as usual, and widens values.Interval.start/end to admit an expression — the class already carried one through get_sub_nodes, its annotation was just narrower than its own API.

Scope — what this does and does not cover

Verified:

• parses, for a property bound, an open bound (".."), and a function bound;
• round-trips through the cql2-json backend:
  to_cql2(parse(q)) → {"op": "t_contains", "args": [{"interval": [{"property": "start_datetime"}, {"property": "end_datetime"}]}, ...]}.

Not covered, and reproducible on this branch: evaluating an interval on the left-hand side. The SQLAlchemy backend's filters.temporal() splits its second argument into low, high and compares a single column against them, so a two-column interval on the left raises

TypeError: '>=' not supported between instances of 'tuple' and 'datetime.datetime'

Fixing that means teaching the backends the interval-vs-interval relations, which is a separate change; I have kept this one to the parsing defect the issue reports. Happy to take that on as a follow-up if you want it.

Tests

4 cases added to tests/parsers/cql2_json/test_parser.py: property/property, property/open, open/property, and a function bound. All 4 fail on the pre-fix code; the module is 45/45 after.

Full suite with the test extra (excluding the modules that need a live Elasticsearch/OpenSearch/Solr or GDAL): 320 passed / 62 skipped / 37 errors against a 316 / 62 / 37 baseline on main — the delta is exactly the 4 new tests; the 37 errors are the network-backed backend suites. ruff / ruff format clean; mypy unchanged from main (same pre-existing values.py shape errors).