[fix](regression) stabilize branch-4.1 colocate column order test

[shuke987]

Proposed changes

Backport the final master-side stabilization for test_colocate_join_of_column_order to branch-4.1.

Master PR #64940 fixed this case by pinning parallel_pipeline_task_num=1 in the three-table colocate explain query. Branch-4.1 still had the earlier stats-injection workaround from #64613, but later failure evidence showed stats pinning was not the root cause: the flaky plan shape is driven by bucket-shuffle downgrade under higher runtime parallelism.

This patch removes the obsolete stats injection and pins parallel_pipeline_task_num=1 for the target explain SQL.

Related to #64960.
Follow-up/backport of #64940.

Testing

• git diff --check
• Not run regression locally; requires Cloud P0-compatible regression environment.

[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 summary for PR #65231

Opinion: I did not find a blocking issue or a substantiated inline comment for this one-file regression-test stabilization.

Critical checkpoints:

• Read the review prompt, review metadata bundle, existing review context, changed-file list, required-agents file, and shared ledger before inspecting the diff.
• No required AGENTS.md files were listed for the changed-file ancestors. The repo-local .claude/skills/code-review/SKILL.md was not present after hidden-path search, so the review proceeded with the bundle instructions and shared-ledger workflow.
• GitHub PR metadata confirms the live head is 754c7efc261eb526d9cc0d83a9b6cbac65071131, with only regression-test/suites/correctness_p0/test_colocate_join_of_column_order.groovy changed.
• Existing inline review context was empty, so no duplicate suppression was needed.
• The patch matches the upstream master stabilization from PR #64940 by pinning parallel_pipeline_task_num=1 for the target explain statement, and this branch additionally removes the older injected-stats workaround.
• The relevant SET_VAR path applies parallel_pipeline_task_num as a statement-scoped session variable, and ConnectContext.getTotalInstanceNum() feeds the bucket-shuffle downgrade threshold that the test is stabilizing.
• After fetching the prompt base SHA, git diff --check passed for the reviewed base/head range and changed file.

Validation limits:

• I did not run the Doris regression suite locally; this case depends on the Cloud P0-compatible regression environment.
• At review time, several GitHub checks were passing, while macOS BE UT, TeamCity compile, and this code-review workflow were still pending.

User focus points:

• No additional user-provided review focus was supplied.

Subagent conclusions:

• optimizer-rewrite: no candidate findings; reviewed optimizer/rewrite, colocate, bucket-shuffle, and semantic-equivalence paths.
• tests-session-config: no candidate findings; reviewed regression-test behavior, expected-output stability, session/config propagation, compatibility, and basic CI/style signals.
• Final convergence round 1 ended with both live subagents replying NO_NEW_VALUABLE_FINDINGS for the same final ledger and empty proposed inline comment set.