Column: make a big slice then divide into sub slices instead of making multi slices in `newColumn`

Enhancement

In some our benchmarks, we find that makeslice of newColumn costs some CPUs

In the newColumn, we can see that we make the slice three times, I think we can only make a big slice instead

I write a simple benchmark

func BenchmarkMakeslice(b *testing.B) {
	cap := 2
	elemLen := 8
	n := 0
	for i := 0; i < b.N; i++ {
		elemBuf := make([]byte, elemLen)
		data := make([]byte, 0, cap*elemLen)
		nullBitmap := make([]byte, 0, (cap+7)>>3)

		n = len(elemBuf) + len(data) + len(nullBitmap) + n
	}
}

func BenchmarkMakeslice2(b *testing.B) {
	cap := 2
	elemLen := 8
	n := 0
	for i := 0; i < b.N; i++ {
		index := cap * elemLen
		cap2 := (cap + 7) >> 3
		buf := make([]byte, elemLen+index+cap2)
		elemBuf := unsafe.Slice(&buf[0], elemLen)
		data := unsafe.Slice(&buf[elemLen], index)
		nullBitmap := unsafe.Slice(&buf[index+elemLen], cap2)
		elemBuf = elemBuf[0:0]
		data = data[0:0]
		nullBitmap = nullBitmap[0:0]

		n = len(elemBuf) + len(data) + len(nullBitmap) + n
	}
}

The result is promising:

cpu: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz
BenchmarkMakeslice-8    	15793832	        84.81 ns/op
BenchmarkMakeslice2-8   	31883019	        37.09 ns/op

Because I think newColumn is frequently used, I guess we can get benefit from this.

BTW, another better way is to use a pool for Column, but seems that we tried this before but failed, maybe we should re-start this again.