…s. Enable new serialization by default for testing

…in MergeTree writer, split dynamic structure and synamic subcolumns logic, split taking dynamic structure and statistics logic

…s-in-map-2

When `optimize_functions_to_subcolumns = 1` and a query has `GROUP BY m`
on a Map column, an expression like `m['key']` in the HAVING clause was
being rewritten to a subcolumn reference `m.key_<key>`. This subcolumn was
not part of GROUP BY, so the analyzer raised an exception.

Root cause: the `always_optimize_identifiers` set (now renamed to
`identifiers_with_filter_optimization`) was populated for any occurrence
of `arrayElement(m, key)` regardless of which clause it appeared in.
This bypassed the "full column also read" guard even for HAVING.

Fix: add clause-context tracking via `in_where_prewhere_stack`. The
`needChildVisit` callback sets the top of the stack to `true` only when
descending into the WHERE or PREWHERE child of a QueryNode. The
`identifiers_with_filter_optimization` set is now populated only when the
occurrence is inside WHERE or PREWHERE. HAVING and other clauses no longer
trigger the exception.

Renamed for clarity:
- `transformers_always_optimize_with_full_column` →
  `transformers_optimize_in_filter_with_full_column`
- `always_optimize_identifiers` →
  `identifiers_with_filter_optimization`

Added regression test: `04040_map_subcolumn_having_bug`.

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>

…a upper bound limit for number of buckets

…s-in-map-2

…s-in-map-2

…s-in-map-2