fix: return flat namespace properties from BigQueryMetastoreCatalog.load_namespace_properties

[anxkhn]

# Rationale for this change

`BigQueryMetastoreCatalog.create_namespace` stores user-supplied namespace
properties in the `parameters` map of `ExternalCatalogDatasetOptions`
(via `_create_external_catalog_dataset_options`). But
`load_namespace_properties` returned
`dataset.external_catalog_dataset_options.to_api_repr()`, which is the nested
BigQuery REST representation: it buries the properties under a `"parameters"`
key and adds a `"defaultStorageLocationUri"` key.

As a result, a simple round-trip does not behave like the other catalogs:

```python
catalog.create_namespace(("db",), {"owner": "alice"})
catalog.load_namespace_properties(("db",))["owner"]  # KeyError
# actually returns {"defaultStorageLocationUri": "...", "parameters": {"owner": "alice"}}

The base MetastoreCatalog contract types this method as -> Properties (a flat
Dict[str, str]), and the Glue and DynamoDB catalogs both return the flat map
(dict(database.get("Parameters", {}))). BigQuery was the outlier. This changes
it to return the flat parameters map, using the same .parameters accessor
already used on the table path in this file. The parameters field is None
when unset, so it is guarded to yield an empty dict for a namespace with no
properties.

Are these changes tested?

Yes. Two regression tests were added to
tests/catalog/test_bigquery_metastore.py:

• test_load_namespace_properties_returns_flat_parameters: creates a namespace
  with {"owner": ..., "comment": ...}, then asserts load_namespace_properties
  returns exactly that flat map (the mock feeds the persisted Dataset back
  through get_dataset). This fails on the old to_api_repr() code (returns the
  nested dict) and passes with the fix.
• test_load_namespace_properties_without_parameters: asserts a dataset whose
  options carry no parameters yields {} rather than None or a nested dict.

python -m pytest tests/catalog/test_bigquery_metastore.py -> 7 passed. Lint
(ruff, ruff-format, mypy, pydocstyle, codespell) passes on both files.
The BigQuery integration tests need Docker + Spark, which were not available in
this environment; the behavior is covered by the unit tests above.

Are there any user-facing changes?

Yes. BigQueryMetastoreCatalog.load_namespace_properties now returns the flat
namespace property map, consistent with the other catalogs, instead of the nested
BigQuery API representation. Code that relied on the previous nested shape
(reading ["parameters"][...] or ["defaultStorageLocationUri"]) would need to
read the properties directly; the previous shape did not match the documented
Properties return type.