Reduce, compare, and apply run 23-26 NIR master darks.

We have been taking pre-run master darks for the NIR detectors, but have not been reducing or applying them. There have been meaningful changes with time, which do show up as residuals from the full cube dark subtraction. Bright pixels themselves are masked, but their neighboring pixels get noticeably over- or under- corrected.

[ This ticket gives the ticket number for the calibration RUNs. We can also put the associated defects (generated from the same input data set) under the same ticket, but that gets strange since defects are curated: we will also get camera re-definitions and the namespace gets confusing. Should we make NIR defects non-curated?]