Fix PhotoCal error calculation to use inverse variance weighting.

Previous versions used the uncertainty as the weight, which is completely
backwards.

python/lsst/pipe/tasks/photoCal.py

@@ -810,8 +810,10 @@ def getZeroPoint(self, src, ref, srcErr=None, zp0=None):
 
         dmag = dmag[good]
         dmagErr = dmagErr[good]
+        zp, weightSum = np.average(dmag, weights=1/dmagErr**2, returned=True)
+        sigma = np.sqrt(1.0/weightSum)
         return pipeBase.Struct(
-            zp = np.average(dmag, weights=dmagErr),
-            sigma = np.std(dmag, ddof=1),
+            zp = zp,
+            sigma = sigma,
             ngood = len(dmag),
         )