Improve performance of min and max aggregation kernels without nulls

[HaoYang670]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.
The function min_max_string does some unnecessary checks when null_count > 0 . For example, we don't need to check has_value in every loop because it will always be true after the first loop.

    if null_count == 0 {
        n = array.value(0);
        for i in 1..data.len() {
            let item = array.value(i);
            if cmp(n, item) {
                n = item;
            }
        }
    } else {
        n = "";
        let mut has_value = false;

        for i in 0..data.len() {
            let item = array.value(i);
            if data.is_valid(i) && (!has_value || cmp(n, item)) {
                has_value = true;
                n = item;
            }
        }
    }

https://github.com/apache/arrow-rs/blob/master/arrow/src/compute/kernels/aggregate.rs#L55-L64

Apart from that, I want this function to be cleaned up because the "for loops" here are not pretty.

Describe the solution you'd like

1. Performance should be improved when null_count > 0
2. No performance penalty is introduced in other cases
3. clean up the code. Maybe use some FP skills

Describe alternatives you've considered
We can also replace array.value(i) by array.value_unchecked(i). But it will introduce some "unsafe", so I am not sure.