Parse unrecognized fields to update offset correctly when deserializing

[tburmeister]

Currently, if we encounter a key in a serialized message with an unknown index, we just ignore it and skip to the next byte, but this is incorrect; we actually need to skip over the size of the field.