Fixes #3326

This PR fixes bug where GROUP BY queries with many aggregate functions fail with "Bytecode is too long" error, and includes significant performance optimizations for the SQL optimizer to speed up query compilation. We had a number of places with O(n²) complexity.

For instance, the following query:

SELECT avg(value), avg(value + 1), avg(value + 2), ... -- 6K functions in total
FROM my_table;

was taking 4 seconds to compile. With this patch, the compilation takes around 300ms.

List of Changes

1. GroupByFunctionsUpdaterFactory Fallback (core fix)

• Added SimpleGroupByFunctionUpdater - a loop-based fallback implementation that avoids JVM bytecode limits
• When aggregate function count exceeds 32, uses the simple loop-based updater instead of generating bytecode
• Threshold determined via JMH benchmarking: bytecode is faster below 32 functions, loop is faster above

2. O(1) Amortized Alias Generation (performance)

• Before: Alias generation was O(n²) - for each new alias, it scanned through sequence numbers 1, 2, 3... until finding an unused one
• After: Added LowerCaseCharSequenceIntHashMap to track the next sequence number for each base alias
• Added aliasSequenceMap to QueryModel (per-model state) and separate maps for pivot/cursor aliases in SqlOptimiser and SqlParser

3. O(n) Duplicate Aggregate Detection (performance)

• Before: O(n²) pairwise compareNodesExact() comparison to detect duplicate aggregates
• After: Hash-based detection using ExpressionNode.deepHashCode() - only compares nodes when hashes collide
• Added detectDuplicateAggregates() method in SqlOptimiser

4. Optimized findColumnByAst (performance)

• Moved findColumnByAst() call inside the if (useOuterModel) check
• Avoids unnecessary O(n) searches when the result won't be used