Set variance in coadd overlaps to mean of the coadd variances

While mosaicking coadds to make templates, use the mean
variances of neighboring coadds to construct the template
variance plane in the patch overlap region. The
coadds have had identical visits contributing to them
and are approximately 100% correlated.

While they may not be exactly 100% correlated (they've had different
backgrounds applied) they are much much closer to 100% correlated
than 0% like calcErrorFromInputVariance assumes.

python/lsst/ip/diffim/getTemplate.py

@@ -571,7 +571,7 @@ def run(self, coaddExposures, bbox, wcs, dataIds, **kwargs):
         statsCtrl = afwMath.StatisticsControl()
         statsCtrl.setNanSafe(True)
         statsCtrl.setWeighted(True)
-        statsCtrl.setCalcErrorFromInputVariance(True)
+        statsCtrl.setCalcErrorMosaicMode(True)
 
         templateExposure = afwImage.ExposureF(finalBBox, finalWcs)
         templateExposure.maskedImage.set(np.nan, afwImage.Mask.getPlaneBitMask("NO_DATA"), np.nan)