Fix seriously misleading documentation in `iris.fileformats.um`

[pp-mo]

The docstrings in iris.fileformats._ff_replacement make it clear that it only applies to "UM files" by which it means Fieldsfiles and LBC
However, these routines are made publicly available in iris.fileformats.um, which refers to "UM files, FieldsFile and PP".

This creates the impression that some or all load routines in iris.fileformats.um -- i.e. 'um_to_pp', 'load_cubes' and 'load_cubes_32bit_ieee') -- will work with either pp or ff, which is not the case.

Needs fixing !

[pp-mo]

I think this explains the confusion seen in the recently-added iris/tests/integration/um/test_fieldsfile.py.
See #2822

[pp-mo]

#2850 fixes ...