I propose reading out the middle couple hundred rows. The wipe for a full frame takes ~7-8s, so this should take ~5s without much effort (i.e. without making a faster wipe).

If we trust the detector map, we could use the first few rows and skip the wipe. So in practice I'd implement this as read window=row0,row1 and wipe up to row0.

Added row0=N nrows=M to ccd read command. This extends the existing nrows= logic, which has always simply read out some rows, without caring about image file format. But if both are specified, a full size image is generated, with the band of rows placed appropriately and the rest left at 0.

For the wipe, requesting nrows=0 makes the exposureState= keywords get spit out correctly and the puts the detector voltages in the integration mode, but no clocking is done. This also extends existing engineering functionality.

When reading a band of rows, W_CDROW0 and W_CDROWN are added to the PHDU.

Still slower than I want, but most of the easy time has been picked out. Umm, for row0=1800 and nrows=400, 11s.

Need to confirm that the data look good, but am waiting on a lamp replacement for that.

Implemented as discussed. Could still make it a bit faster but there is no tall pole.