Select Local Regression Locations Based On Density Rather Than Evenly Spaced

[AyrtonB]

In num_fits_2_reg_anchors need to provide the option to group based on density rather than just evenly spacing them.

In Cleveland 1988, Regression by local fitting: Methods, properties, and computational algorithms (not his more popular paper that was also from 1988 on LOWESS), k-d trees are rather beautifully used to select the locations for n-d smoothing

[AyrtonB]

Good discussion in section 4.2 here - https://www.netlib.org/a/cloess.pdf
And also section 12 here - http://web.ipac.caltech.edu/staff/fmasci/home/astro_refs/LocalRegressionBook_1999.pdf