[fix](regression) Stabilize base compaction file size limit case

[shuke987]

Summary

• Stabilize test_base_compaction_with_dup_key_max_file_size_limit by explicitly setting BE compaction-related configs required by the case.
• Lower base_compaction_dup_key_max_file_size_mbytes to 512 during the test so the loaded rowsets reliably hit the file-size filter.
• Restore original BE config values in finally to avoid polluting the test environment.

Testing

• ./run-regression-test.sh --run -d compaction -s test_base_compaction_with_dup_key_max_file_size_limit
• Remote validation log: /deploy/teamcity_agent13/work/a0477ef061788419/case_output/doris/output/regression-test/log/doris-regression-test.20260708.200842.log
• Result: Test 1 suites, failed 0 suites, fatal 0 scripts, skipped 0 scripts

[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

[shuke987]

skip buildall

[github-actions]

Automated review completed for PR 65386.

No inline issues found in the changed regression test. The patch is limited to snapshotting BE config values, setting the two compaction-related BE configs needed by the case, and restoring the original per-backend values in finally.

Critical checkpoints:

• Changed-file scope: confirmed one modified file, regression-test/suites/compaction/test_base_compaction_with_dup_key_max_file_size_limit.groovy.
• Existing review context: no existing inline review comments or replies were present in .code-review.8OsP7D/pr_review_threads.md.
• Config propagation/restoration: verified get_be_param returns backend-id/value maps and set_original_be_param restores those maps per backend.
• Compaction behavior: verified the lowered base_compaction_dup_key_max_file_size_mbytes still matches the intended base-compaction filter path and final E-808 assertion.
• User focus: no additional user-provided review focus was specified.
• Validation limit: this runner has a shallow/grafted checkout, so I used GitHub PR metadata/patches as authoritative. I did not rerun the heavy p2 regression suite locally; the PR body reports a passing remote run.

Subagent conclusions:

• optimizer-rewrite: no candidates in the initial pass; convergence round 1 returned NO_NEW_VALUABLE_FINDINGS.
• tests-session-config: no candidates in the initial pass; convergence round 1 returned NO_NEW_VALUABLE_FINDINGS.
• Final proposed inline comment set was empty, and convergence round 1 ended with all live subagents reporting NO_NEW_VALUABLE_FINDINGS.

Note: I could not find the repository-local code-review SKILL.md in this checkout, including hidden paths, and no required AGENTS.md files were listed for this PR's changed-file ancestors.