TestRectCentroidFullRange is flaky

[jmr]

=== RUN   TestRectCentroidFullRange
    rect_test.go:1169: [Lo[-90.0000000, -179.9087477], Hi[90.0000000, 179.9087477]].Centroid() - [Lo[-90.0000000, -179.9087477], Hi[90.0000000, -80.4477068]].Centroid() - [Lo[-90.0000000, -80.4477068], Hi[90.0000000, 179.9087477]].Centroid = 1.3322676295501878e-15, want ~0
    rect_test.go:1169: [Lo[-90.0000000, -80.4477068], Hi[90.0000000, 179.9087477]].Centroid() - [Lo[-90.0000000, -80.4477068], Hi[90.0000000, 14.2723948]].Centroid() - [Lo[-90.0000000, 14.2723948], Hi[90.0000000, 179.9087477]].Centroid = 1.0778315928076987e-15, want ~0
--- FAIL: TestRectCentroidFullRange (0.00s)

https://github.com/golang/geo/actions/runs/20675221001/job/59694965751?pr=253#step:5:1032

This was in the context of a PR, but I don't think anything relevant was changed. The number of iterations should be bumped up to try to reproduce this.

geo/s2/rect_test.go

Lines 1168 to 1169 in fd65259

	if got, want := r.Centroid().Sub(child0.Centroid().Vector).Sub(child1.Centroid().Vector).Norm(), 1e-15; got > want {
	t.Errorf("%v.Centroid() - %v.Centroid() - %v.Centroid = %v, want ~0", r, child0, child1, got)

Ah. In C++ the threshold was increased to reduce the flakiness.

https://github.com/google/s2geometry/blob/ac448e2e939863dfefdeb6500ea74fda92d8187c/src/s2/s2latlng_rect_test.cc#L646