Make slit offset measurement more robust

When running reduceArc.py, the results obtained with fitDetectorMap.doSlitOffsets=True are typically much worse than with False. I attribute this to lack of robustness in the slit offset measurement. The SplinedDetectorMap::measureSlitOffsets doesn't do any rejection of outliers. DistortionBasedDetectorMap::measureSlitOffsets uses SplinedDetectorMap::measureSlitOffsets but in the end the slit offsets don't change. Finally, consider moving the measurement of the slit offsets to after the distortion measurement rather than before.