BUG: feed estimated number of distinct values to bloom filter

[dantengsky]

Summary

currently, the "number of distinct values" we feed to the bloom filter is the number of rows, which is too conservative (naive:).
we should use something like hyperloglog to estimate the NDV of give column and then feed the NDV to the bloom filter, for columns that have low cardinality, this will "shrink" the bloom filter index significantly.

https://github.com/datafuselabs/databend/blob/f59efc05bc080e762846fbe8bac24c806e432a75/src/query/storages/index/src/bloom_filter.rs#L151-L162

may also be related to #7314

[BohuTANG]

This may make the bloom filter bitset smaller.
ClickHouse also has a bloom filter index, it seems to work like databend, I haven't checked his bitset size yet.
cc @drmingdrmer

[dantengsky]

vanilla bloom filter is replaced by xor filter in PR #7870