expression: implement vectorized evaluation for builtinRepeatSig

[qw4990]

What problem does this PR solve?

Implement vectorized evaluation for builtinRepeatSig.

What is changed and how it works?

Almost 1.5s faster than before:

BenchmarkVectorizedBuiltinFunc/builtinRepeatSig-VecBuiltinFunc-12                          50000             29567 ns/op
BenchmarkVectorizedBuiltinFunc/builtinRepeatSig-NonVecBuiltinFunc-12                       30000             45092 ns/op

Check List

Tests

• Unit test

[codecov]

Codecov Report

Merging #12014 into master will decrease coverage by 0.0389%.
The diff coverage is 64.1025%.

@@               Coverage Diff               @@
##             master     #12014       +/-   ##
===============================================
- Coverage   81.3794%   81.3405%   -0.039%     
===============================================
  Files           446        447        +1     
  Lines         95658      95646       -12     
===============================================
- Hits          77846      77799       -47     
- Misses        12287      12310       +23     
- Partials       5525       5537       +12

[XuHuaiyu]

LGTM

[SunRunAway]

LGTM

[sre-bot]

/run-all-tests

[zz-jason]

to #12058

[js00070]

i find a TODO at line 34 in expression/builtin_string_vec.go

// TODO: introduce vectorized null-bitmap to speed it up.
if buf.IsNull(i) || buf2.IsNull(i) {
    result.AppendNull()
    continue
}

I'm curious about how this should be done.

Is it right to use function MergeNulls directly? like this

buf.MergeNulls(buf2) // before the for loop
for i := 0; i < n; i++ {
	if buf.IsNull(i) { // reduce a IsNull call here
		result.AppendNull()
		continue
	}

I tried to run benchmark test but didn't find a significant performance improvement.