Make decoding of bytes more robust

[rmeissner]

Some abi encoder encode empty bytes as
0x00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 (length = 0 and an 0-bytes body)

bivrost is expecting 0x0000000000000000000000000000000000000000000000000000000000000000

we make use of this assumption when decoding tuples. Therefore we will have unexpected results if we try to decode tuples that contain the "padded" encoding.

The solution would be to parse the position of each dynamic type and jump to that position for decoding.