Refine the bit_util of Parquet.

[HaoYang670]

Which issue does this PR close?

Closes #1901 .

What changes are included in this PR?

1. Remove the function log2, because:

• The ceiling of log2 (n) == num_required_bits(n - 1)
• log2 is unsafe, which panics when the input number is zero.

2. Change the return type of num_required_bits to u8
3. Use the std function div_ceil
4. optimize trailing_bits

Are there any user-facing changes?

Yes.

[codecov-commenter]

Codecov Report

Merging #1905 (153f4d2) into master (fc4044f) will decrease coverage by 0.01%.
The diff coverage is 94.44%.

@@            Coverage Diff             @@
##           master    #1905      +/-   ##
==========================================
- Coverage   83.43%   83.41%   -0.02%     
==========================================
  Files         203      214      +11     
  Lines       56969    56988      +19     
==========================================
+ Hits        47531    47538       +7     
- Misses       9438     9450      +12     

Impacted Files	Coverage Δ
parquet/src/column/reader.rs	68.61% <0.00%> (ø)
parquet/src/column/reader/decoder.rs	76.27% <100.00%> (ø)
parquet/src/encodings/encoding.rs	93.62% <100.00%> (-0.03%)	⬇️
parquet/src/encodings/levels.rs	93.41% <100.00%> (ø)
parquet/src/encodings/rle.rs	92.55% <100.00%> (ø)
parquet/src/util/bit_util.rs	92.68% <100.00%> (-0.33%)	⬇️
parquet_derive/src/parquet_field.rs	65.75% <0.00%> (-0.46%)	⬇️
arrow/src/datatypes/datatype.rs	65.42% <0.00%> (-0.38%)	⬇️
arrow/src/compute/kernels/temporal.rs	94.64% <0.00%> (ø)
arrow/src/array/builder.rs
... and 13 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fc4044f...153f4d2. Read the comment docs.

[jhorstmann]

This looks good to me, nice improvement!

[tustvold]

I'm not sure if this if block is actually necessary

[HaoYang670]

1<<num_bits would overflow if num_bits >= 64

[tustvold]

Aah yes, forgot you can't just overflow. I presume LLVM is smart enough to turn the if into a mask, i.e.

1 << (num_bits & 64)

[HaoYang670]

Simple test:

    #[test]
    fn test() {
        fn do_test(shift: i32) {
            let i = 1_u64;
            println!("{}", i << shift);
        }
        do_test(65);
    }

Output:

panicked at 'attempt to shift left with overflow'