Add analysis of explain_todo test inconsistencies

Investigation found that the 2446 explain_todo tests contain
fundamental data inconsistencies that prevent systematic fixes:

- LIMIT 0/FORMAT stripping varies between similar tests
- SETTINGS position expectations differ for same syntax
- AND/OR flattening expectations are inconsistent

The root cause is that explain.txt files were generated from
different ClickHouse versions. Fixes that work for explain_todo
tests break existing passing tests.

Detailed analysis in EXPLAIN_TODO_ANALYSIS.md

Author: claude

EXPLAIN_TODO_ANALYSIS.md

@@ -0,0 +1,47 @@
+# EXPLAIN Test Analysis
+
+## Summary
+
+After extensive analysis of the 2446 explain_todo tests, I found that the test data contains fundamental inconsistencies that prevent systematic fixes.
+
+## Key Findings
+
+### 1. LIMIT 0 and FORMAT Stripping
+
+The expected output varies between tests for identical patterns:
+- **02998_system_dns_cache_table**: Expects LIMIT 0 and FORMAT to be STRIPPED
+- **03031_table_function_fuzzquery**: Expects LIMIT 0 and FORMAT to be KEPT
+
+Both tests have `LIMIT 0 FORMAT TSVWithNamesAndTypes;` but expect different outputs.
+
+### 2. SETTINGS Position
+
+The position of `Set` in the output varies:
+- **01293_external_sorting_limit_bug** (explain_todo): Expects Set at SelectQuery level
+- **01104_distributed_numbers_test** (passing): Expects Set at SelectWithUnionQuery level
+
+Changing the logic to fix one breaks the other.
+
+### 3. AND/OR Flattening
+
+Some tests expect flattened boolean operations, others expect nested:
+- **00824_filesystem** (explain_todo): Expects `Function and` with 3 children (flattened)
+- **03653_keeper_histogram_metrics** (passing): Expects nested `Function and` (2 children each)
+
+Implementing flattening broke 173 passing tests.
+
+## Root Cause
+
+The `explain.txt` files were generated from different ClickHouse versions or configurations, leading to inconsistent expected outputs. Without regenerating test data with a consistent ClickHouse version, these inconsistencies cannot be resolved.
+
+## Statistics
+
+- Total tests with explain_todo: 2446
+- Tests with stmt1 in explain_todo: 142
+- Tests currently passing from explain_todo: 0
+
+## Recommendations
+
+1. **Regenerate test data**: Run all tests against a single ClickHouse version to get consistent expected output
+2. **Version-specific logic**: If supporting multiple ClickHouse versions, implement version detection
+3. **Focus on specific patterns**: Fix individual tests rather than broad changes when required