Replace numbers and arrays with LEB128 implementations where appropriate

[Jac0xb]

PR

• Use unsigned LEB128 encoding in conjunction with borsh to reduce instruction data overhead for instructions with offsets and lengths called a CompactU64.

• Use unsigned LEB128 encoding as length for multi-assertion instructions to reduce overhead from 4 bytes per vector to 1 byte (common case) called a LEB128Vec.

• Switch from Option<u16> to CompactU64 for start and length of hash verify for AccountData since offset and length of account data needs to be known in order to hash.