…lickHouse#94017

Add force_primary_key_reverse_order setting that forces all ORDER BY columns
in CREATE TABLE to DESC for stress-testing the reverse key feature across
the entire functional test suite. Auto-enables allow_experimental_reverse_key.

Also fix style check by adding allow_experimental_reverse_key to the
experimental settings ignore list.

Add SET force_primary_key_reverse_order = 0 to tests that are incompatible
with the forced DESC sorting key feature. This includes tests that:
- Check specific ORDER BY / sorting key behavior
- Verify specific data layout or merge behavior
- Use features incompatible with reverse keys (e.g., specific projections)
- Check SHOW CREATE TABLE output

Also register force_primary_key_reverse_order in SettingsChangesHistory.cpp
for version 26.3.

System log tables (query_log, part_log, etc.) are created with
internal=true. Applying force_primary_key_reverse_order to them
produces ORDER BY with DESC keywords that break CI's
setup_log_cluster.sh sed-based ORDER BY modification, causing
syntax errors like 'tuple(event_date DESC, event_time DESC)'.

ATTACH queries load from existing metadata and must not be modified.

Guard the applyForceReverseOrder call with !internal and
mode <= LoadingStrictnessLevel::CREATE.

…m settings

The setting was defaulting to true, which caused ~168 integration test
failures since integration tests create MergeTree tables via normal
CREATE TABLE (not internal). Changing the default to false and adding
the setting to SettingsRandomizer enables random stress-testing in
stateless tests without breaking integration tests.

Tests that are incompatible with the setting already have explicit
SET force_primary_key_reverse_order = 0.

The SET force_primary_key_reverse_order = 0 was placed inside a
Jinja2 list literal within the {% for %} block, breaking template
parsing. Moved it to the top of the file before any Jinja2 syntax.

Three root causes:
- .sh tests: positional SQL args consumed by multitoken() Bool parser;
  fix by using --query flag instead of positional args
- .sql tests with echoOn: inserted SET broke contiguous comment block
  for getTestTagsLength; fix by placing SET after all -- comments
- query_log test: SET was logged; fix by filtering it from output

- 03733, 03533: Update reference files to include leaked comment lines
  (comments after SET break getTestTagsLength contiguity, echoOn echoes them)
- 03006: Move SET before CREATE TABLE so tables aren't affected by random settings
- 02572: Remove ilike filter, add SET entries to reference (SET is logged in query_log)

… tests

Add SET force_primary_key_reverse_order = 0 (or CLICKHOUSE_CLIENT override for
.sh tests) to 54 tests that fail when the setting is randomly enabled via
SettingsRandomizer. Each test has a per-test comment explaining why the setting
is incompatible.

Also fix positional SQL args in 03276 and 03278 backup tests to work around
the multitoken() Bool parsing bug in clickhouse-client.

Add SET force_primary_key_reverse_order = 0 (or CLICKHOUSE_CLIENT override
for .sh tests) to tests that are incompatible with reverse key ordering.
Each test has a one-line comment explaining the specific reason.

…ts and alphabetical order

EXPLAIN PIPELINE output depends on in-order aggregation optimization
which is sensitive to MergeTree key direction.

…rse_order

…isks

# Conflicts:
#	tests/queries/0_stateless/04004_integral_col_comparison_with_float_key_condition.sql

…r upstream merge

…tests

Both tests rely on physical _block_number/_block_offset values from commit
order projections, which change when the sorting key is reversed.

…isks

…r upstream merge