…o query_cache_for_subqueries

…o query_cache_for_subqueries

…_subqueries

…display

…ckHouse into query_cache_for_subqueries

`finalizeWriteInQueryResultCache` is a no-op when the pipeline has no
cache writers, so the conditional checks are unnecessary. Making the
calls unconditional is consistent with `executeQuery.cpp`'s
`finalizeQueryPipelineBeforeLogging` which always calls it, and is
more robust for future changes.

Remove the now-unused `shouldCacheScalarSubquery` function and
`query_cache_for_subqueries` extern declarations from both
`ExecuteScalarSubqueriesVisitor.cpp` and `PreparedSets.cpp`.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add null guard for `query_result_cache` in `Planner::buildPlanForQueryNode`
to prevent crash when a subquery explicitly sets `use_query_cache = true`
but the server has no query cache configured.

Remove spurious duplicate `optimize_and_compare_chain` entry in
`SettingsChangesHistory.cpp` that was introduced during rebase.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

`finalizeWriteInQueryResultCache` is a no-op when the pipeline has no
cache writers, so the conditional checks are unnecessary. Making the
calls unconditional is consistent with `executeQuery.cpp`'s
`finalizeQueryPipelineBeforeLogging` which always calls it, and is
more robust for future changes.

Remove the now-unused `shouldCacheScalarSubquery` function and
`query_cache_for_subqueries` extern declarations from both
`ExecuteScalarSubqueriesVisitor.cpp` and `PreparedSets.cpp`.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Eliminate double AST clone in `QueryResultCache::Key` constructor:
  introduce `calculateAstHashAndQueryString` that clones once for
  subqueries and computes both hash and query string from the result.
  Add `already_cloned` parameter to `calculateAstHash` to skip its
  internal clone when called with a pre-cloned AST.

- Guard `extremes`/`max_result_bytes`/`max_result_rows` default
  normalization with `if (is_subquery)` so non-subquery cache entries
  are not affected (avoids hash changes and cache invalidation on
  upgrade for existing non-subquery entries).

- Add tests for CTE subquery caching, UNION ALL subquery caching,
  explicit per-subquery opt-in with cache hit verification, old
  analyzer negative test, and `query_cache_min_query_runs` with
  subqueries.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The AI reviewer correctly identified that `setCanUseQueryResultCache(true)`
on the planner context leaks subquery opt-in into the shared query state,
which could cause outer query results to be cached even when the outer
query didn't set `use_query_cache`.

Fix: instead of mutating the shared context, add a `skip_context_check`
parameter to `checkCanWriteQueryResultCache` that bypasses the context
flag check while still enforcing safety checks (nondeterministic
functions, system tables). The subquery cache eligibility is now tracked
locally via `local_can_use_cache` without mutating global state.

Also adds Test 14: regression test proving that explicit subquery opt-in
does NOT create a top-level cache entry when outer `use_query_cache = 0`.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Upstream master bumped to 26.4, so `query_cache_for_subqueries` setting
history entry must be in the 26.4 section (was in 26.3 which is now closed).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Revert contrib/NuRaft, contrib/libstemmer_c, contrib/rust_vendor and
.gitignore to upstream master — these were accidentally included from
the upstream merge and are not part of this feature.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

- Initialize `chunk_type` in `StreamInQueryResultCacheStep` to prevent
  potential use of uninitialized variable if `StreamType` enum is extended
- Fix `system.query_cache` column description for `is_subquery`
- Fix minor grammar in comment
- Add missing trailing newlines in test files

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Resolve conflict in Planner.cpp: keep the local_can_use_cache
approach (from reviewer feedback) instead of mutating the shared
query context.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…opagation