Sort by detector id to ensure reproducibility

Order of inputs is not guaranteed, and the weighting per detector to the
full focal plane sky model can depend (very weakly) on the order of inputs,
so sort them by detector to guarantee reproducible results for the same set
of inputs.

python/lsst/pipe/drivers/skyCorrection.py

@@ -201,8 +201,13 @@ class SkyCorrectionTask(pipeBase.PipelineTask, BatchPoolTask):
 
     def runQuantum(self, butlerQC, inputRefs, outputRefs):
 
-        # reorder skyCalibs and calBkgArray per calExpArray
+        # Reorder the skyCalibs, calBkgArray, and calExpArray inputRefs and the
+        # skyCorr outputRef sorted by detector id to ensure reproducibility.
         detectorOrder = [ref.dataId['detector'] for ref in inputRefs.calExpArray]
+        detectorOrder.sort()
+        inputRefs.calExpArray = reorderAndPadList(inputRefs.calExpArray,
+                                                  [ref.dataId['detector'] for ref in inputRefs.calExpArray],
+                                                  detectorOrder)
         inputRefs.skyCalibs = reorderAndPadList(inputRefs.skyCalibs,
                                                 [ref.dataId['detector'] for ref in inputRefs.skyCalibs],
                                                 detectorOrder)
@@ -286,6 +291,9 @@ def runDataRef(self, expRef):
 
             dataIdList = [ccdRef.dataId for ccdRef in expRef.subItems("ccd") if
                           ccdRef.datasetExists(self.config.calexpType)]
+            ccdList = [dataId["ccd"] for dataId in dataIdList]
+            # Order inputs sorted by detector id to ensure reproducibility
+            dataIdList = [dataId for _, dataId in sorted(zip(ccdList, dataIdList))]
 
             exposures = pool.map(self.loadImage, dataIdList)
             if DEBUG: