Antalya 25.8 Cherry-pick: Force mutation to recalculate indices when rewriting compact parts

[mkmkme]

Cherry-pick: Force mutation to recalculate indices when rewriting compact parts

Same as #2043, but for Antalya 25.8

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Fix "TOO_MANY_MARKS" error which could have happened after some ALTER queries for compact parts (ClickHouse#91980 by @alesapin)

Documentation entry for user-facing changes

...

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Performance tests
• Aarch64 tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• All Regression
• Disable CI Cache

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• OAuth (5m)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

[mkmkme]

Tested locally and confirmed that the added test fails without the change and passes with the change. No review or verification needed, will wait for CI though

[github-actions]

Workflow [PR], commit [3a4c6e1]

[mkmkme]

@blau-ai

[blau-ai]

CI triage for #2162

Verdict: 3 failing checks, 0 PR-caused. 2 are external-suite/version-skew (Iceberg), 1 is a CI-reporting infra glitch (FinishCIReport). Nothing here is caused by this backport — the diff touches only src/Storages/MergeTree/MutateTask.cpp + one new stateless test and contains no Iceberg code. All checks that actually exercise the changed code are green.

---

❌ Regression x86 iceberg_1 / RegressionTestsRelease / Iceberg (1) — NOT PR-caused (external suite vs binary skew)

❌ Regression x86 iceberg_2 / RegressionTestsRelease / Iceberg (2) — NOT PR-caused (external suite vs binary skew)

Decisive log line (iceberg_2, /iceberg/iceberg engine/glue catalog/position delete reads and /iceberg table engine/feature/{position delete,compaction} smoke):

Code: 552. DB::Exception: Unrecognized option '--allow_insert_into_iceberg'. (UNRECOGNIZED_ARGUMENTS)

The clickhouse-regression suite is pulled at HEAD (not pinned to this PR) and now passes --allow_insert_into_iceberg to the server. The 25.8 binary doesn't have that Iceberg-write option, so those scenarios abort at setup — this is a suite-vs-binary version mismatch, not a code regression. iceberg_1 shows the same shape (40 ok, 1 failed, 3 errored; the "errored" scenarios are Iceberg-write/glue setup, not query correctness).

Evidence it's not this PR:

• The diff is a MergeTree mutation fix (force index recalculation when rewriting compact parts); it never touches Iceberg, S3, or catalog code.
• The same Iceberg (1)/Iceberg (2) checks are green on the base branch antalya-25.8, so it's not something this backport introduced.

Next step: safe to ignore / re-run; it will keep failing until either the regression suite is pinned to a 25.8-compatible ref or the --allow_insert_into_iceberg Iceberg-write feature is backported to 25.8. This is a suite-compat issue for the regression-infra owners, not a fix for this PR.

---

❌ FinishCIReport — NOT PR-caused (reporting infra, 404)

Decisive log lines:

+ REPORT_LINK='WARNING:Failed to fetch workflow config from
  https://altinity-build-artifacts.s3.amazonaws.com/PRs/2162/<sha>/config_workflow/workflow_config_pr.json: 404 ...
+ echo 'Error: WARNING:Failed to fetch workflow config ...'
+ exit 1
##[error]Process completed with exit code 1.

The final report step couldn't fetch workflow_config_pr.json from S3 (404) and exits 1. Note the Config Workflow check itself passed, so this is a transient artifact-availability/reporting glitch, unrelated to the code change.

Next step: re-run the job; it should clear once the artifact is present.

---

PR health check

The change itself looks healthy. Every check that exercises the modified code path is green: Fast test (0 failed), all Stateless matrices (debug/binary, parallel & sequential, s3/DatabaseReplicated/ParallelReplicas — 0 failed), all Integration tests (0 fail), Stress, AST fuzzer (amd_debug), BuzzHouse, and every Build. Crucially, the fix's own regression test 03755_ttl_materialize_bug ships in this diff and passed in the stateless runs — so the TOO_MANY_MARKS scenario it guards against is covered. Recommend re-running FinishCIReport and treating the Iceberg reds as a known regression-suite compatibility issue rather than a merge blocker for this backport.

— @blau-ai (automated CI triage; I can't build/run ClickHouse locally, so this is evidence-based from CI logs.)