Add dedicated cas suite to regression and split lightweight_delete CAS jobs

[alsugiliazova]

Add the dedicated cas / cas_s3_cache suite from clickhouse-regression to the Antalya 26.6 pipeline, gated on the same ci_regression_cas checkbox as the other CAS jobs.

Match the clickhouse-regression release CAS job splits:

• lightweight_delete: four shards (disk space, hard restart, leftover, and the slow features), each with a 5-hour timeout.
• attach partition: three parts (part 1, part 2, part 3).

Related: #2234

Changelog category (leave one):

• CI Fix or Improvement (changelog entry is not required)

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

...

Documentation entry for user-facing changes

• I will add a documentation PR to ClickHouse/clickhouse-docs right after this PR is merged.
• I need help to write the documentation.
• No documentation needed.

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Unit 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)
• CAS (content-addressed storage; Antalya only)
• 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)

[github-actions]

Workflow [PR], commit [f14cedc]

[CarlosFelipeOR]

I addressed some fixes/improvements:

1 — Selects suite

cas_selects was failing on /selects/final/force/concurrent, not because of this PR: the CAS disk exhausts the container's ephemeral ports against the store and the mount lease gets fenced — #2243, 15 of 47 job executions since 2026-08-26. Fixed in clickhouse-regression by setting http_keep_alive_timeout 60s and http_keep_alive_max_requests 1000 on the CAS disk (69be3aa72), in the shared helper so it covers every --cas suite. Rerun is green, 3020/3020, after three failed attempts on the same build.

Failing job: https://github.com/Altinity/ClickHouse/actions/runs/34214700789/job/102257899198

2 — Engines suite

cas_engines was failing on /engines/replicated_replacing_merge_tree/without is deleted, not because of this PR: the scenario reads the second replica right after inserting on the first, with no SYSTEM SYNC REPLICA. Under CAS the replica fetch takes ~267 ms against ~11 ms on local disk, so the SELECT lands before the part is committed and returns 0 rows — 9 failures in 109 CAS runs since 2026-08-18, 0 in 590 non-CAS runs. Fixed in clickhouse-regression by syncing the replica before reading it (3830fc45e), applied to all five scenarios of the feature rather than just the failing one — Altinity/clickhouse-regression#165. 15 of 15 passes on the build that failed 13 of 16 before. Rerun is green, 37/37, after two failed attempts on the same build.

Failing job: https://github.com/Altinity/ClickHouse/actions/runs/34214700789/job/102128879154

3 — CAS jobs were overwriting each other's S3 reports

engines, cas_engines and cas_s3_cache_engines all uploaded to the same prefix, and so did every other CAS/non-CAS pair: 38 of the 98 concrete jobs shared 11 prefixes. setup.sh already builds a /cas or /cas-s3-cache suffix, but it reads $regression_args, which this workflow never exposed to the job environment — the input only becomes REGRESSION_ARGS in a step that runs after setup.sh, so the suffix was always empty. A job retry overwrote the failing attempt too, which is why the server logs of a CAS failure were routinely lost and report_url in gh-data.clickhouse_regression_results pointed at the wrong job. Fixed here in 0fa0a34 by exposing regression_args to setup.sh, plus giving CASAlter/CASS3CacheAlter the storage_path the non-CAS Alter job already has (their replace and move variants share suite=alter with an empty part, so the suffix alone would not separate them). Companion fix in clickhouse-regression (a0e488a1): the guard that skips the suffix compared against STORAGE=/cas while tiered_storage_cas actually sets /with_cas. 99 jobs now resolve to 99 distinct prefixes.

[CarlosFelipeOR]

LGTM