Stable 25.8: stabilize Stress test (JOIN chain + concat nullability LOGICAL_ERRORs)

[CarlosFelipeOR]

Backport of :

1. Disable join_use_nulls for 03716_multiple_joins_using_top_level ClickHouse/ClickHouse#91128 - PR: disable join_use_nulls for 03716_multiple_joins_using_top_level_identifier
2. PR: fix n-way join with GLOBAL join ClickHouse/ClickHouse#86895 - PR: fix n-way join with GLOBAL join
3. PR: fix LEFT/INNER ... RIGHT ... JOINS chain ClickHouse/ClickHouse#87178 - PR: fix LEFT/INNER ... RIGHT ... JOINS chain
4. PR: fix RIGHT joins chain ClickHouse/ClickHouse#97316 - PR: fix RIGHT joins chain

Includes a follow-up fix to address issues introduced during conflict resolution and align the implementation with upstream.

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):

Backport upstream fixes for LOGICAL_ERROR in queries with parallel replicas and multiple JOINs (LEFT/INNER...RIGHT, RIGHT...RIGHT, n-way with GLOBAL/FULL) (ClickHouse#86895, ClickHouse#87178, ClickHouse#97316 by @devcrafter)

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)

[github-actions]

Workflow [PR], commit [fb8a546]

[CarlosFelipeOR]

QA Verification

Verdict: ✅ Approved — no regressions. The only outstanding CI failure is a pre-existing flaky test unrelated to this PR's changes.

Failing test

04043_system_asynchronous_inserts_user_filter — job Stateless tests (amd_tsan, s3 storage, parallel).

This is a timing flake in the test itself, not a regression:

• Non-deterministic on the identical commit (fb8a546): the same test in the same job produced FAIL → OK → FAIL. A real regression would fail deterministically.
• Only fails in the single heaviest config. It passes in all ~15 other configurations (debug, asan, msan, ubsan, AsyncInsert, ParallelReplicas, distributed plan, etc.), failing only under TSAN + S3 + parallel — the slowest, most timing-sensitive environment.
• Failure is a count mismatch, not a crash: secret_user/admin expected 1, saw 0. The test does an async INSERT with a 600 s flush timeout and expects the entry to stay pending in system.asynchronous_inserts while it is queried; under TSAN+S3 the entry was no longer queued at query time. The isolation logic actually under test (restricted_user = 0) behaved correctly.
• No connection to the changes. This PR only touches the JOIN planner (PlannerJoinTree.cpp), JOIN test files, and parallel_replicas_blacklist.txt. The failing test contains no JOINs and was itself recently added by an unrelated backport (upstream Fix cross-user data leak in system.asynchronous_inserts ClickHouse/ClickHouse#100024, "Fix cross-user data leak in system.asynchronous_inserts").

Stress test (amd_tsan)

Since this PR targets stress-test stability, Stress test (amd_tsan) was run 4 times on the final commit fb8a546 — all 4 passed (Not crashed, No logical errors, No sanitizer asserts, No queries hung, Server started/died checks all OK). Confirmed stable.

[mkmkme]

This has been handled in Antalya 25.8 by #1724. No need to backport