Use the number of input points as the estimated number of
points when setting up the point locator for the contour
filters, rather than the cell-based estimate used previously.

On a wavelet data set with 1M points and 1M hex cells, the
locator ended up holding 1.5M points -- far exceeding the
estimated 62K points. This caused bucket sizes to grow to
~120 points per bucket in cases, while the locator only expects
around 3 points per bucket. This resulted in a ~10% speedup
of the filter updates.

Change-Id: I3fde70306b046d9f3c79a3fbc4cfbd5c3a230028