Improve RLP macros

Our RLP heavily depends on Ethereum's implementation(https://github.com/paritytech/rlp).
They provide some procedural macros that generate encoding/decoding code for basic objects,
but it lacks of many details and functionality.
Though it helps on some level, we had to manually implement encoding/decoding routines in many cases.
I believe that there will be countless new types of data that should be serialized, and we are going to code repetitive functions again and again if there's no some new powerful metaprogramming support.

(see https://github.com/CodeChain-io/foundry/issues/65)