[fix](parquet) Isolate Variant planning from ordinary scans

[Gabriel39]

What

• Cache whether a Parquet schema subtree contains Variant data.
• Build Variant physical types and materialization plans only for Variant-containing columns.
• Keep ordinary Parquet scans on their existing logical-type path.

Why

Variant planning was performed for every projected column and row group, even when the schema contained no Variant data. This added repeated schema traversal and allocation overhead to ordinary scans.

Validation

• Cherry-picked cleanly onto the latest branch-4.1.
• The resulting patch is equivalent to the original isolated fix.
• Local tests were not run for this cherry-pick-only PR.

[hello-stephen]

Thank you for your contribution to Apache Doris.
Don't know what should be done next? See How to process your PR.

Please clearly describe your PR:

1. What problem was fixed (it's best to include specific error reporting information). How it was fixed.
2. Which behaviors were modified. What was the previous behavior, what is it now, why was it modified, and what possible impacts might there be.
3. What features were added. Why was this function added?
4. Which code was refactored and why was this part of the code refactored?
5. Which functions were optimized and what is the difference before and after the optimization?

[Gabriel39]

run buildall

[Gabriel39]

/review

[github-actions]

Automated review: changes requested. One P2 performance regression remains: a partial complex projection that excludes all Variant descendants is still treated as Variant-bearing because the new gates test plan-pointer presence rather than the projection-sensitive plan content. Schema propagation, logical/physical type ownership, file-context/footer-cache lifecycle, row-group and shredded-statistics conservatism, row shape/cursors, and external compatibility otherwise checked out. The added tests validate the cached booleans but do not exercise the Variant-excluding projection. Review was static as required; repository builds/tests were not run. No additional user focus was provided, so the full PR received the normal review. Review-cycle status: all full-coverage and risk-focused agents converged on this frozen candidate set; landed-state verification follows submission.

[github-actions]

[P2] Keep the Variant gate projection-sensitive

contains_variant describes the complete schema subtree, not the selected projection. For a reachable STRUCT<label, info VARIANT> scan that projects only label, build_variant_plan() follows the projection and returns contains_variant == false, but the plan pointer is still non-null. This check therefore sends every read/select and accumulated lazy-skip batch through _variant_physical_column, recursive wrapper transformation/shape validation, and replacement scratch allocation even though no Variant node was requested; the base path decoded directly because it checked the plan's content. Please discard a plan whose contains_variant is false (or preserve the content-sensitive gates), and cover a Variant-excluding partial projection.

[github-actions]

PR approved by at least one committer and no changes requested.

[hello-stephen]

BE UT Coverage Report

Increment line coverage 100.00% (43/43) 🎉

Increment coverage report
Complete coverage report

Category	Coverage
Function Coverage	58.29% (24579/42164)
Line Coverage	42.47% (246903/581382)
Region Coverage	38.56% (195076/505857)
Branch Coverage	39.86% (89086/223512)

1

[hello-stephen]

BE Regression && UT Coverage Report

Increment line coverage 100.00% (43/43) 🎉

Increment coverage report
Complete coverage report

Category	Coverage
Function Coverage	73.50% (30265/41178)
Line Coverage	57.42% (331985/578177)
Region Coverage	54.33% (276426/508758)
Branch Coverage	55.25% (123544/223615)