Precalculate values inside HashTableGrower

[nickitat]

Changelog category (leave one):

• Performance Improvement

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Optimized insertion and lookups in the HashTable.

[nickitat]

perf tests: 1 3 4

perf tests with alignment: 1 3 4

there are few slowdowns caused apparently by increasing grower size:

0.271	0.338	+1.246x	0.246	0.246	functions_with_hash_tables	4	select groupUniqArray(rand() % 100) from numbers(1000 * 1000) group by number / 1000 format Null

1.194	1.383	+1.157x	0.157	0.157	uniq	11	SELECT ClientIP AS k, uniqExact(UserID) FROM hits_100m_single GROUP BY k FORMAT Null

both use HT with stack memory with initial_size_degree = 4. In theory we could provide another HashTableGrower implementation for them, but IMO it doesn't really makes sense 'cause queries are fairly artificial.

[nickitat]

new perf tests: 1 3

[nickitat]

new perf tests: 1 2 + a couple of queries with dictionaries also became slightly faster.

slowdown in single_fixed_string_groupby #0 doesn't reproduce. slowdowns in other queries don't reproduce either and moreover they shouldn't use hash tables at all.

[alexey-milovidov]

Let's check what's wrong with WITH FILL test.
It is definitely unrelated... but we can postpone merging this PR one or two days just to eliminate another flaky test more quickly.

[alexey-milovidov]

LGTM

[nickitat]

Let's check what's wrong with WITH FILL test.

I'm trying to fix it here #38705