branch-4.0: [fix](test) deflake several branch-4.0 P2 regression cases

[morningman]

What problem does this PR solve?

Deflake several pre-existing flaky/failing cases in the branch-4.0 P2 regression suite (surfaced by an internal P2 run). None of these are caused by product behavior changes.

1. opensky_p2 count / avgDistance / totalDistance / mostBusyOrigin — backport of [fix](regression) fix p2 opensky failure #62447: the S3 source csv.gz was actually a tar.gz, so after it was corrected the loaded row count changed (+30 rows) while the committed .out still held the old values. Update the four .out files and re-enable the NumberTotalRows == NumberLoadedRows load assertion.

2. segcompaction_p2/test_segcompaction_agg_keys — backport of [fix](regression) Fix case test_segcompaction_agg_keys #61026: the AGGREGATE REPLACE winner for a duplicate key is not deterministic. S3 load parallelizes into segments whose boundaries are non-deterministic (adaptive memtable flush / memory pressure) and the read-merge keeps the first row by segment order, so which of the duplicate col_0=47 rows wins varies between runs. Assert exactly one row plus the value being one of the legitimate outcomes, and delete the brittle .out.

3. inverted_index_p2/test_show_data (test_show_data_with_compaction) — it compared index sizes with inverted_index_compaction_enable on vs off for exact equality. For identical data the on-disk size can differ slightly due to file/merge layout, so compare within a 10% tolerance instead, which still catches gross index bloat or corruption.

4. compaction/test_base_compaction_with_dup_key_max_file_size_limit — the test builds a specific rowset layout via manual compactions and expects a manual base compaction to be rejected with E-808 (input rowset exceeds base_compaction_dup_key_max_file_size_mbytes). When the BE global disable_auto_compaction is false, background compaction races the manual steps and reshapes the rowsets, and an in-test short-circuit masked the result. Disable auto compaction cluster-wide for the duration of the test (restored in finally) and assert on the real response.

5. compaction/test_single_replica_compaction — the waitForCompaction and getTabletStatus closures assigned code/out without def, which the regression framework's script-source guard rejects (defined global variables in script are not allowed: code). [fix](case) fix global vars used in cases #55933 declared only process; declare the remaining locals too.

Release note

None

🤖 Generated with Claude Code

https://claude.ai/code/session_01QuX5zypeHzV1ZgT9o8BzoF

[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?

[morningman]

run buildall

[github-actions]

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

[github-actions]

PR approved by anyone and no changes requested.