[branch-4.1][fix](test) backport check_before_quit variant round-trip fix

[shuke987]

Summary

Cherry-pick the master fix from #65093 to branch-4.1.

• Pin default_variant_enable_doc_mode=false before check_before_quit recreates tables from SHOW CREATE TABLE.
• Keep schema-change columns on the legacy test_predefine1 flatten-nested table in non-doc mode, so the table left by variant_p0/doc_mode/predefine/load.groovy remains recreate-able by SHOW CREATE TABLE.

branch-4.1 already had the older #64613 pinning for max subcolumns and sparse hash shard count. This backport adds the missing doc-mode pin and the follow-up flatten-nested table fix.

Testing

• git diff --check origin/branch-4.1...HEAD
• Not run locally. check_before_quit validates the state left by a full regression pipeline.

[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?

[shuke987]

run buildall

[shuke987]

/review

[github-actions]

Automated review completed for PR #65228.

I reviewed the two changed regression scripts, the GitHub patch metadata, existing inline review context, and the relevant variant session-default/parser/type rendering paths. I did not find a substantiated issue that needs an inline review comment.

Critical checkpoints:

• Changed files: checked check_before_quit.groovy and variant_p0/doc_mode/predefine/load.groovy.
• Duplicate context: no existing inline review comments or replies were present.
• Variant default pinning: the new doc-mode pin matches the existing round-trip check intent; the other actually randomized variant defaults in initFuzzyModeVariables() are already pinned by the surrounding block.
• Doc-mode schema-change columns: the explicit variant_enable_doc_mode=false properties are accepted by the parser, do not combine with mutually exclusive variant properties, and keep the legacy flatten-nested table consistent with the existing expected-output shape.
• User focus: no additional user-provided focus was supplied.
• Repo skill note: the requested repo-local code-review SKILL.md was not present in this checkout after explicit search, so I proceeded with the prompt workflow and shared-ledger/subagent process.
• Validation limits: static review only. The local checkout does not contain the PR base commit, so I used GitHub PR patch/file metadata as the diff authority; I did not run the regression suites.

Subagent conclusions:

• optimizer-rewrite: no candidates in the initial pass; convergence round 1 returned NO_NEW_VALUABLE_FINDINGS.
• tests-session-config: proposed TSC-001 about default_variant_enable_nested_group; I dismissed it with evidence because that variable is not randomized by SessionVariable.initFuzzyModeVariables() and VariableMgr has no separate reflective fuzzy randomizer. Convergence round 1 returned NO_NEW_VALUABLE_FINDINGS.

Overall opinion: no blocking correctness or test-stability issue found in this PR.