du: our implementation is slower than GNU's

[sylvestre]

On firefox source code (lot of files and directories):

# with 0.0.17:

$ hyperfine "/usr/bin/du " "/usr/lib/cargo/bin/coreutils/du" 
Benchmark 1: /usr/bin/du
  Time (mean ± σ):      3.804 s ±  0.051 s    [User: 1.174 s, System: 2.606 s]
  Range (min … max):    3.722 s …  3.884 s    10 runs

Benchmark 2: /usr/lib/cargo/bin/coreutils/du
  Time (mean ± σ):      4.822 s ±  0.056 s    [User: 1.608 s, System: 3.190 s]
  Range (min … max):    4.754 s …  4.921 s    10 runs

Summary
  '/usr/bin/du ' ran
    1.27 ± 0.02 times faster than '/usr/lib/cargo/bin/coreutils/du '

seems that we regressed:

# with 0.0.26:

hyperfine "/usr/bin/du " "/usr/lib/cargo/bin/coreutils/du "
Benchmark 1: /usr/bin/du
  Time (mean ± σ):      3.980 s ±  0.075 s    [User: 1.181 s, System: 2.772 s]
  Range (min … max):    3.887 s …  4.087 s    10 runs

Benchmark 2: /usr/lib/cargo/bin/coreutils/du
  Time (mean ± σ):      6.947 s ±  0.096 s    [User: 2.533 s, System: 5.479 s]
  Range (min … max):    6.800 s …  7.104 s    10 runs

Summary
  '/usr/bin/du ' ran
    1.75 ± 0.04 times faster than '/usr/lib/cargo/bin/coreutils/du '

[sylvestre]

@jesseschalken has been working on it

#6839
#6840
#6841

[sylvestre]

With the three PR, we are almost the same level:

$ hyperfine "/usr/bin/du" "/home/sylvestre/dev/debian/coreutils/target/release/coreutils du"
Benchmark 1: /usr/bin/du
  Time (mean ± σ):      4.108 s ±  0.113 s    [User: 1.237 s, System: 2.843 s]
  Range (min … max):    3.885 s …  4.312 s    10 runs

Benchmark 2: /home/sylvestre/dev/debian/coreutils/target/release/coreutils du
  Time (mean ± σ):      4.256 s ±  0.073 s    [User: 1.954 s, System: 3.315 s]
  Range (min … max):    4.100 s …  4.333 s    10 runs

Summary
  '/usr/bin/du' ran
    1.04 ± 0.03 times faster than '/home/sylvestre/dev/debian/coreutils/target/release/coreutils du'

[sylvestre]

a profile with the 3 PR https://share.firefox.dev/4fdZpVX