It sounds quite non-academic to me to apply an encoding-correcting
view on a memory-mapped range and therefore apply the endian
correction
only to the parts of the range I actually end up traversing.

I was under the impression that people typically use memory mapping to map e.g. a file onto a buffer of bytes, in which case being able to traverse a byte array with reversed endianness would provide more utility than just the quasi-transform-view that wraps byteswap.