imprint_subract renaming combined with fits_hash results in de-sync of ASDF and FITS

[braingram]

#775 renamed meta.cal_step.imprint to meta.cal_step.imprint_subtract both using the S_IMPRNT keyword. This can result in desync of the ASDF and FITS data when opening a file written prior to that PR. If we assume the file hasn't been modified in some way to invalidate the FITS_HASH (the user hasn't manually modified the fits contents) opening the file will:

• skip mapping FITS to ASDF (since the hash matches)
• the datamodel will have a meta.cal_step.imprint value that matches the value of S_IMPRINT (since the hash check didn't fail)
• the datamodel will not have a meta.cal_step.imprint_subtract

However if the model is modified to invalidate the hash opening the file will:

• map FITS to ASDF
• meta.cal_step.imprint will have a value matching the prior S_IMPRINT value
• meta.cal_step.imprint_subtract will have a value matching the current S_IMPRINT value

Are there any places where the pipeline inspects meta.cal_step.imprint_subtract outside of test code? If not, this is a likely inconsequential issue.

FWIW the issue was found when testing different options for mapping FITS to ASDF contents. I think this issue is also further motivation for removing the use of FITS_HASH (which might be practical with the improvements to this mapping). However removing FITS_HASH will be complicated by #381 as noted in #271