[fix](regression) Use single bucket for unique key delete compaction cases

[shuke987]

Proposed changes

Set an explicit single bucket for the unique-key-with-delete compaction cases:

• test_compaction_uniq_keys_with_delete.groovy
• test_compaction_uniq_keys_with_delete_ck.groovy

Both cases keep the original strict rowCount < 8 * replicaNum check and the final qt_select_default3 query result check after cumulative compaction.

Why

Cloud P0 build 987265 failed in compaction.test_compaction_uniq_keys_with_delete.test_compaction_uniq_keys_with_delete:

assert (rowCount < 8 * replicaNum)
8 | 8 1
false

The failing build already included the table-level "disable_auto_compaction" = "true" change from #65211. The build log shows cumulative compaction completed with [OK], but the active rowsets can still contain exactly 8 * replicaNum physical rows for this unique-key + delete scenario.

The default bucket layout lets the same case observe different physical row-count distributions across tablets. This patch fixes the test table to BUCKETS 1, so the strict cumulative-compaction row-count assertion is checked under a single-tablet layout instead of depending on cross-tablet physical distribution.

Validation

• git diff --check
• Inspected TeamCity build 987265 compressed build log:
  • table DDL includes "disable_auto_compaction" = "true"
  • cumulative compaction status is [OK]
  • active row-count sum is exactly 8, matching 8 * replicaNum
• Checked the final PR diff against the PR merge base; it only adds BUCKETS 1 to the two case tables.

Buildall requested after the bucket-only update.

Behavior changed

No user-facing behavior change. Regression case deflake only.

[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]

run buildall

[hello-stephen]

/review

[github-actions]

PR approved by anyone and no changes requested.

[github-actions]

Automated review summary for PR #65352.

I reviewed the final PR-scoped patch from GitHub: it changes only two compaction regression suites and adds BUCKETS 1 to each unique-key-with-delete test table. I did not find a substantiated issue to raise as an inline comment.

Critical checkpoint conclusions:

• Goal and proof: The PR goal is a regression-test deflake for the unique-key delete compaction row-count check. Pinning the table layout to one bucket addresses the cross-tablet physical row-count variability while preserving the strict rowCount < 8 * replicaNum assertion and the final ordered result checks.
• Scope: The change is minimal and test-only: two DDL additions, no expected-output or production-code changes.
• Concurrency/lifecycle: No new runtime concurrency or lifecycle behavior is introduced. The existing helper still triggers/waits compaction per tablet.
• Config/session/compatibility: No new config item, session variable, persisted format, FE/BE protocol, or rolling-upgrade compatibility path is introduced.
• Parallel paths: The non-cluster-key and cluster-key variants are updated consistently.
• Test coverage/results: The changed files are regression tests, the generated .out files remain valid because query output is unchanged and ordered, and git diff --check passed for the PR-scoped changed files. I did not run the regression suite locally because this checkout lacks .worktree_initialized and thirdparty/installed; review validation was static plus GitHub PR metadata/CI state.
• Observability, transactions, persistence, data writes, and performance: Not applicable beyond the existing test behavior, because this is a bucket-count pin in regression DDL only.

Subagent conclusions:

• optimizer-rewrite: no candidates found; convergence round 1 returned NO_NEW_VALUABLE_FINDINGS.
• tests-session-config: no candidates found; convergence round 1 returned NO_NEW_VALUABLE_FINDINGS.
• Existing inline review context was empty, so there were no duplicate threads to suppress.

User focus: No additional user-provided review focus was supplied.