It seems more natural to treat the "subversion" field (=user agent string, see BIP 14) of a node as pure string rather than a bytestring within the test framework. This is also suggested with the naming prefix in msg_version.strSubVer: one probably wouldn't expect a field starting with "str" to be a bytestring that needs further decoding to be useful. This PR moves the encoding/decoding parts to the serialization/deserialization routines so that the user doesn't have to bother with that anymore.
This is a backport of core#20993