…uckdb#23375)

Resolved (rerere-applied, matching the full-merge resolution):
    test/sql/error/test_try_expression.test_slow

This correctly allows lambda parameters to be referenced inside the
`try` expression.

…uckdb#23375)

Brings this change from the v1.5-variegata release branch. It overlapped
with the main/v2.0 line; the conflicts were resolved as follows:

- `test/sql/error/test_try_expression.test_slow`: kept both main's RESET ieee_floating_point_ops and variegata's lambda-parameters-inside-TRY test cases.

…uckdb#23375)

Brings this change from the v1.5-variegata release branch; it overlapped with
the main/v2.0 line and was resolved as follows:

- `test/sql/error/test_try_expression.test_slow`: kept both main's RESET ieee_floating_point_ops and variegata's lambda-in-TRY test cases.

Merge the v1.5-variegata release branch (323 commits since the 40721d5
merge-base) into the main/v2.0 line. ~55 files conflicted, almost entirely
because release-branch backports re-landed fixes that main had implemented
independently (often against a newer API), plus a few genuine feature/security
additions only on the release branch.

Conflict resolutions (grouped):

Already-in-main (took main, dropped the duplicate backport):
- external_file_cache refactor: main split the monolithic files into
  external_file_cache/; variegata's memleak/generation/disablement commits are
  the same upstream work (PR duckdb#22857, duckdb#22961) against the old layout -> dropped.
- autocomplete transform_expression.cpp: the empty-subscript guard was promoted
  into the core PEG parser (src/parser/peg/.../transform_expression.cpp) -> dropped.
- query_profiler.cpp: main re-architected the profiler; the null-deref fix
  variegata added to the old switch already exists in RenderProfilingNodeTree.
- list_select.cpp: the OOB child_idx guard already exists in main.
- bind_pivot.cpp duckdb#21931: main's restructured ExtractUnpivotEntries already
  guards against binding a non-existent expr.
- bind_operator_expression.cpp duckdb#23375: main reuses the current binder via an
  inside_try flag, so lambda bindings are already in scope inside TRY(...);
  variegata's TryOperatorBinder + explicit lambda propagation is unnecessary.
  (variegata's lambda-in-TRY tests are kept to prove this.)
- arrow_type_extension.cpp Bool8: main's VectorIterator/Writer already resolve
  dict/constant/sliced sources via ToUnifiedFormat.

Ported variegata's fix into main's newer API:
- geometry_stats.cpp: ported strip_geometry_cast (CRS-erasing cast look-through)
  to GetChildren()/GetReturnType()/GetValue(); CanPruneXY relax-pruning merged.
- alprd_scan.hpp / rle.cpp: ported variegata's out-of-bounds-read hardening
  (bounds checks + IOExceptions) onto main's GetDatabase()/GetBlockHandle()/
  GetDataMutable() API and member-pointer RLE scan state.
- variant_binary_decoder.cpp: kept variegata's bounds-checked
  (data, offset, data_size) decode signatures (the merged header), folded in
  main's TIMESTAMPTZNS handling and includes.
- multi_file_column_mapper.cpp duckdb#23354: kept main's static MapColumnMap
  (MultiFileLocalIndex) and added variegata's IsInvalidMapKeyDefault NULL-MAP-key
  guard.
- json: kept main's 3-field DefaultMacro layout + json_copy_strftime macros, but
  folded in variegata's "reject NULL json key" behavior (json_group_object macro
  + AddKeyValuePairs throw).

Took main's newer architecture (variegata's old-API version superseded):
- physical_table_scan.cpp: main's TableFunctionGetMetricsInput get_metrics.
- parquet_reader.hpp/.cpp: main's row_groups_read/skipped + prefetch cost model
  (subsumes variegata's row_groups_scanned); resolved a diff3-garbled hunk.
- parquet_writer.cpp: main's options/transform handling within variegata's
  MatchesTypes two-loop InitializePreprocessing structure.
- date_part.cpp: main's executor-count-removal kept; took variegata's 3-param
  PropagateSimpleDatePartStatistics<MIN,MAX,T> calls (matches merged definition).
- struct_data.cpp: variegata's AppendChild (adopted across the merged appenders).

Config / version data (union or main):
- version_map.json + regenerated storage_info: union of v1.5.4 (68/7) and
  v2.0.0 (69/8); latest/default follow v2.0.
- extension .cmake pins: kept main's v2.0-compatible tags + APPLY_PATCHES.
- force_storage.json: kept main's sections, added variegata's geometry_shred_more.
- NotifyExternalRepositories.yml / enable_verification_for_debug.json: honored
  main's deletion/restructuring.

Tests: kept variegata's lambda-in-TRY cases, PIVOT special-key serialization,
and PRAGMA enable_verification additions.

--- Compile & behaviour adaptations (so the merge builds and the full suite is green) ---

- Parquet metrics (duckdb#23140): exposed variegata's row-group scan metrics through
  main's TableFunctionGetMetricsInput API (single ParquetScanGetMetrics reading
  row_groups_read/skipped); dropped the duplicate old-API ParquetGetMetrics.
- geometry_stats.cpp (duckdb#23140): looked through CRS-erasing GEOMETRY casts via
  BoundCastExpression::Child() (the child member is private on main).
- geometry_parquet_pushdown.test (duckdb#23140): ported to main's profiler API
  (SET tracked_metrics with dotted names + JSON navigation), asserting the
  deterministic row_groups_scanned (total_row_groups_to_scan is a racy per-thread count).
- parquet_writer.cpp (duckdb#23234): bound the transform-data re-initialization to
  main's ParquetWriterOptions::sql_types (options.sql_types).
- Arrow AppendChild (duckdb#23190): AppendChild() takes a const Vector& and References
  it into a mutable local before the extension's duckdb_to_arrow cast, so the
  struct/array appenders can pass main's const children.
- pivotref.cpp (duckdb#23162): serialized PIVOT GROUP BY via
  WriteOptionallyQuoted(groups[i].GetIdentifierName()) for main's Identifier type.
- date_part.cpp (duckdb#23254): kept variegata's infinity-aware
  PropagateSimpleDatePartStatistics body but used main's authoritative
  second/minute/week/ms/us/ns upper bounds (59/53/...), not the leap-second 60/54.
- alprd_corrupted_dict_size.test (duckdb#23100): main's stricter exceptions_count guard
  fires first for the corrupt fixture, so the test now expects main's
  "ALPRD exceptions_count exceeds vector size" message (both guards kept).

…uckdb#23375)

Brings this change from the v1.5-variegata release branch; it overlapped with
the main/v2.0 line and was resolved as follows:

- `test/sql/error/test_try_expression.test_slow`: kept both main's RESET ieee_floating_point_ops and variegata's lambda-in-TRY test cases.

This correctly allows lambda parameters to be referenced inside the
`try` expression.

Upstream-Commit: duckdb/duckdb@c4770ec