header-only implementation of intersection

[isVoid]

Implement header only API for intersection_count and intersection.

template<typename Range1, typename Range2, typename OutputIt>
OutputIt pairwise_linestring_intersection(Range1 multilinestrings1, Range2 multilinestring2, OutputIt output_begin, stream);

template<typename Range1, typename Range2, typename OutputIt>
OutputIt pairwise_linestring_intersection_count(Range1 multilinestring1, Range2 multilinestring2, OutputIt output_begin, stream);

Depending on #763 and #764. With the help of these primitive and data structures, I posturizes that the kernel should be parallelized on 1-segment per thread and iterate on the multilinestring in the other multilinestring.

Kernel pseudo code without needing to write indices

__global__ void intersection_count_kernel(range1, range2, output):
	tidx = get_threadidx()
	Segment seg1 = range1.segment(tidx)
	For (mls : range2)
		For (seg2 : mls)
			Point_opt, line_opt = segment_intersection(seg1, seg2)
			If (point_opt != nullopt)
				atomicInc(Output[tidx].num_intersecting_points)
			If (line_opt != nullopt)
				atomicInc(Output[tidx].num_overlapping_segments)

__global__ void intersection_kernel(range1, range2, temp_storage, output):
	tidx = get_threadidx()
	Segment seg1 = range1.segment(tidx)
	For (mls : range2)
		For (seg2 : mls)
			point_opt, line_opt = segment_intersection(seg1, seg2)
			If (point_opt != nullopt)
				Npoint = atomicReadAndInc(temp_storage.npoint)
				Output[npoint] = point_opt.value()
			If (line_opt != nullopt)
				NLine  = atomicReadAndInc(temp_storage.nline)
				Output[nLine] = line_opt.value()

[UPDATE: 11-8]
count API is only counting the upper bound of the number of intersection/overlaps - it does not give any promises to the number of results. So count will not be exposed as a public API.

#765 (comment)