flash: Further optimize depth computations

[sampsyo]

This adds two new optimizations:

• When doing interval depth, avoid needlessly computing unique depth. That took us from roughly 1.3s to roughly 720ms.
• When using depth just to measure path length, add an optimization that skips the depth computation altogether. That took us from there to about 400ms.

Altogether, these optimizations are worth a speedup of about 3.2×.

Checking in on all our options for doing this one weird computation:

Benchmark 1: sh odgi_wdepth.sh > odgi_result.bed
  Time (mean ± σ):     15.008 s ±  0.119 s    [User: 13.973 s, System: 2.919 s]
  Range (min … max):   14.920 s … 15.144 s    3 runs

Benchmark 2: fgfa -i chr8.flatgfa window-depth "chm13#chr8" 5000 > fgfa_result.bed
  Time (mean ± σ):     972.5 ms ±   2.5 ms    [User: 844.5 ms, System: 125.5 ms]
  Range (min … max):   970.1 ms … 975.0 ms    3 runs

Benchmark 3: flash flash_wdepth.sh > flash_result.bed
  Time (mean ± σ):     710.0 ms ±   3.5 ms    [User: 502.4 ms, System: 206.2 ms]
  Range (min … max):   706.6 ms … 713.6 ms    3 runs

Benchmark 4: flash -O flash_wdepth.sh > flash_result.bed
  Time (mean ± σ):     394.0 ms ±   2.1 ms    [User: 271.7 ms, System: 120.6 ms]
  Range (min … max):   391.8 ms … 396.0 ms    3 runs

Summary
  flash -O flash_wdepth.sh > flash_result.bed ran
    1.80 ± 0.01 times faster than flash flash_wdepth.sh > flash_result.bed
    2.47 ± 0.01 times faster than fgfa -i chr8.flatgfa window-depth "chm13#chr8" 5000 > fgfa_result.bed
   38.09 ± 0.36 times faster than sh odgi_wdepth.sh > odgi_result.bed

So on the bottom line, we are now about 38× faster than odgi.