About the stability of COPY BINARY data

From https://www.postgresql.org/docs/current/sql-copy.html:
|> binary-format file is less portable across machine architectures
and PostgreSQL versions

In my experience, the binary encoding of binding/resultset/copy is
endian neutral (network byte order), so what is the less portable
across machine architectures that warning about?

Also, does the code for per-type _send() and _recv() functions really change
across versions of PostgreSQL? How common are instances of such
changes across versions? Any examples of such backward-incompatible
changes, in the past?

The binary data contains OIDs, but if sticking to built-in types,
which OIDs are unlikely to change across versions?

I'm obviously storing COPY BINARY data (we have lots of bytea
columns), and I wonder how bad it is long term, and across PostgreSQL
versions.

Thanks for any insights, --DD