fix: improvements and fixes to unnest support

[k-anshul]

1. unnest did not work for snowflake and bigquery.
2. Filter on unnest exploded the list for databricks.

Follow ups:

1. Subquery with unnest does not work on main.

Checklist:

• Covered by tests
• Ran it and it works as intended
• Reviewed the diff before requesting a review
• Checked for unhandled edge cases
• Linked the issues it closes
• Checked if the docs need to be updated. If so, create a separate Linear DOCS issue
• Intend to cherry-pick into the release branch
• I'm proud of this work!

[nishantmonu51]

One note that does not anchor to a changed line. The Databricks behaviour change is worth a release-note item: a Subquery filter on an unnest dimension that is not selected — what the explore UI generates for measure-based dimension filters — now fails with the right value must be a list of values for an array IN condition (runtime/metricsview/astexpr.go:684) instead of running. It previously ran, but through the row-exploding LATERAL VIEW join this PR removes, so the numbers were double-counted. Turning silently wrong results into an error is the right trade, but Databricks users will experience it as a regression until the subquery follow-up lands.

[nishantmonu51]

The unconditional skip means none of the new Snowflake SQL is exercised against a real database — not in this PR's checks, and not after the account is reactivated, since nothing about reactivation removes a hard-coded t.Skip. That leaves the LATERAL FLATTEN(...) t0 (seq, key, path, index, tags, this) column-alias form and the FILTER(..., t0 -> ...) lambda with bound ? parameters verified only by local runs. testruntime.AcquireConnector does not skip on missing credentials on its own — it fails via require.NotEmpty (runtime/testruntime/connectors.go:80) — so skipping on an empty RILL_RUNTIME_SNOWFLAKE_TEST_DSN here would let the test come back by itself.

[k-anshul]

DSNs are set but the accounts are inactive. Same convention is being followed in all the test cases.

[nishantmonu51]

Same as the Snowflake case: the hard-coded skip leaves the EXISTS(arr, t0 -> ...) lambda, the arrays_overlap path and the switch to tupleStyle: true in LateralUnnest unverified against a real warehouse, and it will outlive the inactive account. Gating on an empty RILL_RUNTIME_DATABRICKS_TEST_DSN instead of an unconditional skip would restore coverage automatically.

[k-anshul]

DSNs are set but the accounts are inactive. Same convention is being followed in all the test cases.

[k-anshul]

One note that does not anchor to a changed line. The Databricks behaviour change is worth a release-note item: a Subquery filter on an unnest dimension that is not selected — what the explore UI generates for measure-based dimension filters — now fails with the right value must be a list of values for an array IN condition (runtime/metricsview/astexpr.go:684) instead of running. It previously ran, but through the row-exploding LATERAL VIEW join this PR removes, so the numbers were double-counted. Turning silently wrong results into an error is the right trade, but Databricks users will experience it as a regression until the subquery follow-up lands.

It is okay to fix it correctly later. No need for release note item since measure filters with unnest cols are rare.