Drifter slows when many depos are given

[brettviren]

When run inside larsoft, G4 causes on order 600k depos to be fed to WCT simulation. This leads to an ~hours/event of running time. A simple test that mostly just drifts:

Some doubling:

• 5591 depos 1.2 s (0.2 mm)
• 11181 depos 3.2 s (0.1 mm)
• 22361 depos 11.5 s (0.05 mm)
• 44722 depos 44.5 s (0.025 mm)

Roughly O(N^2)....

The problem is in the loop in flush_ripe(). Trial fix is to change xr.depos into a sorted set and then quit from the loop as soon as now is reached.

[brettviren]

After trial fix:

• 5591 depos 0.46 s
• 11181 depos 0.54 s
• 22361 depos 0.61 s
• 44722 depos 0.76 s
• 89444 depos 1.2s (2 tracks, 2ms apart)
• 2236068 depos 18.8s (2 tracks, 0.001mm steps)

[brettviren]

FYI, this bug fix can be exercised with:

wire-cell -c cfg/pgrapher/test/wct-test-drifter.jsonnet

Then, change the step size in the Jsonnet.