Define binary representation for TypeIDs and implement binary marshaling

[loreto]

As suggested in #4, typeids could implement the encoding.BinaryMarshaler and encoding.BinaryUnmarshaler interfaces.

To do so, we would first have to define in the spec the standard binary representation for typeids. Options include:

• Re-use the string representation and encoded in UTF-8
• Instead of encoding the prefix as the bytes of a base32 string, be more efficient by using the "raw" 128-bit representation of the UUID. A follow up question would be how to distinguish between the prefix and suffix fields in the binary representation (use a separator? include an integer indicating the length of the prefix?)