If you want to provide a spec for encoding public keys as octet
streams containing sequences of lines delimited by line termination
sequences, that's fine, but that's (a) less useful (because it
requires converting between octet-stream representations and native
representations for filesystems that don't use line termination
sequences) and (b) not what publickeyfile-08 calls itself.
Um, section 2 makes it fairly clear that the format described is for
exchanging public keys between implementations, rather than
necessarily for use within an implementation. It could probably be
clearer.
It's clear that's what motivated it. It's not clear that it's intended
to be restricted to that.
I'm also not convinced that it makes any difference. A text file is a
reasonable unit for interchange; there is no need to reduce it to an
octet sequence of defined contents. After all, such interchange may
well be between implementations running on the same OS or even the same
machine. While most storage mechanisms do reduce a text file to an
octet sequence at some point, that octet sequence does not need to fall
under the jurisdiction of the spec - nor would it be particularly
useful for it to do so.