Compare with the usual rules for protobuf evolution:

All fields are optional. This is because fields have a lifecycle. Your code might be reading data generated by some other code written before the field was defined, or generated by some other code written after the field became obsolete.

With protobufs, you can never reuse a field number. However, you can remove a field (keeping its number reserved), which effectively means ignoring it in new applications. It's neither read nor written, which is fine because see above.

You can also rename the field itself. This doesn't affect the wire protocol.

It seems more sensible than having misleading names around for people to trip over. Maybe we'd be better off if databases used field numbers to refer to columns in the wire protocol?