Gets closer to GNU ls, ~8% gain vs current main.

cargo build -r -p uu_ls && taskset -c 0 hyperfine --warmup 100 -L ls ls,target/release/ls,./ls-main "{ls} -lR /var/lib .git || true"
Benchmark 1: ls -lR /var/lib .git || true
  Time (mean ± σ):      27.2 ms ±   0.6 ms    [User: 11.9 ms, System: 15.0 ms]
  Range (min … max):    26.5 ms …  28.6 ms    109 runs
 
Benchmark 2: target/release/ls -lR /var/lib .git || true
  Time (mean ± σ):      27.9 ms ±   0.9 ms    [User: 12.2 ms, System: 15.2 ms]
  Range (min … max):    26.9 ms …  35.2 ms    104 runs

Benchmark 3: ./ls-main -lR /var/lib .git || true
  Time (mean ± σ):      30.2 ms ±   1.4 ms    [User: 14.2 ms, System: 15.4 ms]
  Range (min … max):    29.3 ms …  42.4 ms    96 runs

Summary
  ls -lR /var/lib .git || true ran
    1.02 ± 0.04 times faster than target/release/ls -lR /var/lib .git || true
    1.11 ± 0.05 times faster than ./ls-main -lR /var/lib .git || true

ls: Optimize time formatting

Instead of recreating the formatter over and over again, keep
it pre-parsed in a variable in TimeStyler class.

Also, avoid calling now over and over again, that's also slow.

Improves performance by about 6%.

ls: display_uname/group: Return a reference

Cache even numerical strings (numeric_uid_gid) in the HashMap,
this makes very little difference performance wise.

However, this allows us to return a reference to a String instead
of making a clone.

Saves about 2-3% on ls -lR /var/lib .git (and ls -lRn).

Also, add a note that HashMap might not be the most optimal choice.

ls: ListState: Add uid/gid cache to the structure

Easier to reason about than the LazyLock/Mutex encapsulated
static variables.

Performance difference is not measurable, but this drops uneeded
Mutex lock/unlock that were seen in samply output.

ls: Create a ListState struct to maintain state

We put the out writer and style manager in there, for now.

Reduces the number of parameters to pass around, and we'll add
more useful things in there.

Little to no performance difference.