Fix clang 23 -Wlifetime-safety errors

[azat]

Fix:

• -Wlifetime-safety-dangling-global-moved in ReplicasReconnector,
• -Wlifetime-safety-dangling-field-moved in MergeTreeReaderStream and TTLTransform
• -Wlifetime-safety-return-stack-addr-moved in BackupsWorker (move first, alias later)
• -Wunused-but-set-variable for inside_main in keeper.
• -Wlifetime-safety-use-after-scope-moved in ProfileEvents

Disable -Wlifetime-safety-invalidation: too many false positives, it treats mutation of a container captured by reference in a lambda as invalidating the reference to the container variable itself.

Changelog category (leave one):

• Not for changelog (changelog entry is not required)

[clickhouse-gh]

Workflow [PR], commit [b9d2d91]

Summary: ❌

• Performance Comparison: Performance dashboard

job_name	test_name	status	info	comment
Stateless tests (arm_asan_ubsan, targeted)		FAIL
	Server died	FAIL	cidb
	Scraping system tables	FAIL	cidb
AST fuzzer (amd_debug, targeted, old_compatibility)		FAIL
	Logical error: Sort order of blocks violated for column number A, left: B, right: C. Chunk D, rows read E.F (STID: 3413-350b)	FAIL	cidb
Stateless tests (amd_asan_ubsan, db disk, distributed plan, sequential)		FAIL
	03631_hive_columns_not_in_format_header	ERROR	cidb
	Server died	FAIL	cidb
	Segmentation fault (STID: 3973-48d2)	FAIL	cidb

---

AI Review

Summary

This PR fixes Clang 23 warning cleanup in the warning configuration and in pointer/ownership assignment order for ReplicasReconnector, MergeTreeReaderStream, TTLTransform, BackupsWorker, and ProfileEvents; keeper_main now matches the existing external inside_main pattern from the other entry points. I found no review-blocking correctness issues in the changed code, and Not for changelog is appropriate for this compiler-cleanup-only PR.

Missing context / blind spots

• ⚠️ The available Praktika report for commit b9d2d91a currently has no test results, and GitHub still shows Build (arm_tidy) and Fast test in progress; green CI, especially the Clang 23/tidy build that exercises these warnings, would close the build-evidence gap.

Final Verdict

Status: ✅ Approve. No code changes requested; merge should still wait for the pending CI jobs that validate the compiler-warning contract.

[clickhouse-gh]

LLVM Coverage Report

Metric	Baseline	Current	Δ
Lines	84.40%	84.40%	+0.00%
Functions	91.40%	91.40%	+0.00%
Branches	77.00%	77.00%	+0.00%

Changed lines: Changed C/C++ lines covered by tests: 35/38 (92.11%) | Lost baseline coverage (was covered on master, now uncovered in this PR): 2 line(s) · Uncovered code

Full report · Diff report

[alexey-milovidov]

it treats mutation of a container captured by reference in a lambda as invalidating the reference to the container variable itself

We can annotate these cases in a subsequent PR and enable the warnings. I don't mind doing it in 1000 places :)

[azat]

We can annotate these cases in a subsequent PR and enable the warnings. I don't mind doing it in 1000 places :)

There are not too many (AFAFIR around 63), but not sure that it worth it, since looks like it is a bug in the clang analysis