Finalize protobuf encodings

[cwgoes]

Such that all encoded data in the provable store has a canonical encoding defined by the spec.

[mossid]

Some opinions:

• If a key-value element has a primitive type, it does not need to be encoded as a message(no tags).
• For numbers, fixed{32, 64} should be preferred(nonsigned nonvarint little-endian). varints should be avoided.
• bytes(ASCII) should be preferred over string(UTF-8).
• bytes are prepended with varint encoded length, which is unnecessary for our use case. We should prefer to encode without length prefix whenever possible.
• bools are safe to use.

In conclusion, the preferred set of types is {fixed64, bytes, bool}, without tags if possible.

[cwgoes]

We should prefer to encode without length prefix whenever possible.

Why? Merkle proofs, for example, often won't be constant-length.

[mossid]

Only when there is no ambiguation, for example, simple bytes can be encoded without prefix when it is not in a message(as it cannot be ambiguous), but should be encoded with prefix when it is in a message.

For example:

message M {
  bytes F1 = 1;
  bytes F2 = 2;
}

F1 and F2 should be encoded with prefix(because there are two bytes and omitting length prefix will ambiguate them), but when the value is not a message and a single bytes, it does not need to be defined as a message so the length prefix can be omitted

[cwgoes]

I see what you mean; in the single-bytes case I agree that we should not length prefix.

[cwgoes]

Adding a document describing the canonical encoding to the spec.