That is so that decode·encode becomes identity.

For the other way

	encode(decode(pickle)) == pickle

it is generally not possible to do, because pickle machine is almost
general machine, and it is possible to e.g. have some prefix in program
which is NOP or result of which is no longer used, use different opcodes
to build list or tuple etc.

Adjustments to encoder are:

- teach it to encode big.Int back to opLong (not some struct)
- teach it to encode Call back to opReduce (not some struct)

Tests are adjusted to verify that `decode(encode(v)) == v` holds for
all inputs in test vector.