[SPARK-56596][SQL] Enable dual runs for single-pass analyzer

[mihailotim-db]

What changes were proposed in this pull request?

This PR enables single-pass analyzer dual runs in testing mode and adds the infrastructure to control dual-run behavior.

Background: How dual runs work

The HybridAnalyzer sits between the SQL parser and the rest of the query planning pipeline. When dual-run mode is enabled, every query goes through
the following flow:

1. The ResolverGuard inspects the unresolved logical plan to check whether all operators and expressions are supported by the single-pass resolver.
   If any unsupported feature is detected, the query skips dual-run entirely and uses the legacy fixed-point analyzer only.
2. If the guard passes, the fixed-point analyzer runs first and produces a resolved plan. During this run, relation metadata (catalog lookups, temp
   view resolutions) is captured in the AnalyzerBridgeState.
3. The single-pass resolver then runs on the same unresolved plan. It uses the BridgedRelationMetadataProvider to reuse the relation metadata from
   step 2, avoiding duplicate catalog RPCs.
4. The results are compared:
   • If both succeed, their output schemas and normalized logical plans are compared. A mismatch indicates a bug in the single-pass resolver.
   • If the single-pass fails but the fixed-point succeeds, the behavior depends on ANALYZER_LOG_ERRORS_INSTEAD_OF_THROWING_IN_DUAL_RUNS: in tests
     it throws (to catch regressions), in production it logs and returns the fixed-point result.
   • If the single-pass throws ExplicitlyUnsupportedResolverFeature, the fixed-point result is returned silently (this is expected for features not
     yet ported to the resolver).
   • The fixed-point result is always returned as the final plan (ANALYZER_DUAL_RUN_RETURN_SINGLE_PASS_RESULT defaults to false).

Why are the changes needed?

Dual runs are the primary mechanism for validating single-pass analyzer correctness against the legacy fixed-point analyzer. Enabling them by default
in tests ensures that any regressions in single-pass resolution are caught immediately during development, rather than discovered later in production.

Does this PR introduce any user-facing change?

No. All changes are internal and only affect test-time behavior. The ANALYZER_LOG_ERRORS_INSTEAD_OF_THROWING_IN_DUAL_RUNS config defaults to true
outside of tests, so production behavior is unchanged.

How was this patch tested?

Existing tests (SQLQueryTestSuite and others). The dual-run mode validates single-pass results against fixed-point results on every query executed
during tests.

Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (claude-opus-4-6)

[dongjoon-hyun]

Please create a JIRA issue and use it in the PR title before converting this back to a normal PR.