Fix Delaunay robustness by using longer floating point variable

[Komzpa]

GEOS ticket: https://trac.osgeo.org/geos/ticket/922#ticket

PostGIS ticket: https://trac.osgeo.org/postgis/ticket/4152

[dbaston]

Long double is 64 bits in MSVC; I think you need to ifdef out the test.

[Komzpa]

@dbaston is there a chance to get it into 3.7.0 and backpatch to 3.6 before release of 3.7? :)

[dbaston]

I think so. You said on the JTS ticket that you had reduced the failure case to a 5-point input; can you use that here?

[dbaston]

@Komzpa ^^^

[Komzpa]

@dbaston using smaller case now, passes locally

[dbaston]

@Komzpa Great! Can you please squash the commits and force-push?

[Komzpa]

done. isn't there squash-and-merge button on github PR merge though, why is it manual?

[dbaston]

This is just a mirror of the GEOS repository.