Chipper has pathological query pattern

[hobu]

With the work in #159, the chipper now has to do dimension lookups every time for every point because it reads one point at a time.

[c42f]

Doesn't the chipper just call copyPointsFast() when it's doing the one-at-a-time thing? If I'm reading things correctly that shouldn't involve dimension lookup.

I think there's a different problem. If I run valgrind against pcpipeline test/perf/chipper.xml (which includes a filters.cache larger than all the points in autzen-thin.las), I still get 10653 calls recorded against pdal::drivers::las::PointDimensions::PointDimensions() - one for each point. Shouldn't the cache just read them in once?

Not sure I understand what's going on here :)

Here's what I'm using for profiling.

1. Create a release build with debug info turned on
2. Run valgrind on test/perf/chipper.xml as follows:
   valgrind --tool=callgrind --simulate-cache=yes pcpipeline chipper.xml
3. Check out the results in kcachegrind