The SkyWcs version of lsst.meas.astrom.rotateWcsPixelsBy90 on
DM-10765 works by inserting a pixel transform at the start of the transform chain. That is simple and robust. However, it means that AST does not know how to write the result out to FITS metadata as a TAN-SIP WCS.
One obvious alternative is to rotate the WCS the same way the old code did: by extracting the SIP terms, rotating them and creating a new WCS. The problem with that approach is that it is specific to TAN-SIP WCS and it relies on AST's ability to reliably write out TAN-SIP. If AST fails to do that for some reason (e.g.
DM-13539) then the rotated WCS will not operate as expected.
I think the long-range alternative (if we still need to rotate WCS after fitting them with meas_mosaic or jointcal) is to rotate the WCS as mentioned at the beginning, then fit a SIP to that when writing FITS metadata, as enabled by
DM-13065. However, I hope we can deal with that after merging DM-10765.
For now I am not sure what to do. I have asked David Berry if he has an suggestion, but I think it is a stretch to expect AST to be able to recognize the rotated WCS as TAN-SIP for purpose of writing FITS-WCS header cards.