perf: stream analyze statistics with KLL sketch

[KKould]

Summary

• add a KLL sketch builder/view for approximate quantile buckets
• stream ANALYZE statistics into KLL and Count-Min Sketch instead of fully sorting all values
• use a shared relative error setting for histogram and CM sketch statistics
• refine index scan costing for unique equality and ordered index hints

Analyze Benchmark

Release builds comparing origin/main (b5965dc) against this PR after replacing memcomparable sorting with direct DataValue comparison.

where_by_index_explain.slt, five runs, measuring only analyze table t1:

version	avg	median	min	max
main	25.09ms	24.92ms	24.77ms	25.70ms
current	13.28ms	13.28ms	13.10ms	13.47ms

TPCC with heaptrack, --backend kitesql-lmdb --measure-time 1 --num-ware 1:

version	wall	max RSS	analyze sum
main	31.02s	667,136 KB	1.46s
current	31.10s	254,552 KB	1.13s

Heaptrack peak memory consumers:

version	largest peak
main	402.65M
current	9.82M

Tests

• make test-all
• cargo fmt --check
• cargo check -p kite_sql