General science pipeline optimizations

After profiling the science pipeline (4 exposures of the same field), I found the following potential opportunities for optimizing a percent or so of runtime here and there:

• Disable aperture corrections and photometry until we actually need them.
• Implement CentroidLinesTask.makeCatalog in C++.
• Implement our own argminmax in C++ (with NAN-safe option) for H4RG ISR (instead of calling np.argmin and np.argmax separately).
• Fix asPfsFiberArraySet to avoid Table::__iter__.
• Optimize what we can in fitContinuum: (binData function, change use of np.percentile to our quartiles, conversion of std::vector to numpy).