feat: support geospatial results on the kernel path

[cathleeny]

Summary

• add the kernel-only, default-enabled enable_geospatial_support connection option
• return canonical {"srid": int, "wkb": bytes} values when enabled and WKT / EWKT strings when disabled
• preserve logical GEOMETRY / GEOGRAPHY Arrow metadata
• pass the value through the typed PyO3 argument only; it is never forwarded to SEA

Dependency

Depends on databricks/databricks-sql-kernel#361

Validation

• unit suite: 1,023 passed, 4 skipped, 2 deselected
• real-kernel signature/routing tests against the exact kernel extension: 2 passed
• live Reyden E2E against the exact extension: default-enabled, explicit-enabled, and disabled-string modes passed through both direct Arrow-scalar and pandas row conversion paths, including Arrow metadata, WKB equality, and nulls

[aakash-saravanan-db]

Could we raise the optional databricks-sql-kernel dependency minimum to the first published wheel that accepts enable_geospatial_support before this ships? pyproject.toml still allows ^1.1.0, while this keyword is checked on every kernel-backed open_session, even for queries with no geo columns. If a user upgrades the Python driver but retains an allowed older kernel wheel, would every kernel connection fail with NotSupportedError? Could we cover the minimum-supported-wheel installation in a compatibility test?

[aakash-saravanan-db]

Python-side implementation looks good. Please resolve the existing minimum-supported Kernel wheel/dependency-floor comment before merging. The additional durable E2E coverage suggestion is nonblocking.