Update Poseidon to last changes of the reference implementation from
26ddaa91db

Compatible with update at circomlib
(cf853c1cc9)

Since Poseidon Hash is used because of compatibility in zkSNARK circuits, due
circuit constraints number, the hash method of [T]*big.Int is the one directly
compatible with the circuits, is the method which have the `Hash` name on it.
The method that can take arbitrary length of []*big.Int putting them in chunks
of [T]*big.Int and iterating, is called `HashSlice`. The `HashBytes` has been
removed, as is a method that will not be used in zkSNARK circuits due high
constraints number.

For zkSNARK circuits, should be used `poseidon.Hash([poseidon.T]*big.Int)`.

Optimize MiMC7 migrating from *big.Int to goff generated finite field
operations.

There is still a lot of room for optimization for MiMC7 in the way that is done internally, but will be done in the future.

Benchmarks:
Tested on a Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz, with 16GB of RAM.

- Before:
```
BenchmarkMIMC7-4   	    1026	   1160298 ns/op
```

- After this commit:
```
BenchmarkMIMC7-4   	   19263	     61651 ns/op
```

Optimize Poseidon migrating from *big.Int to goff generated finite field
operations.

Benchmarks:
Tested on a Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz, with 16GB of RAM.

- Before the optimizations:
```
BenchmarkPoseidon-4                  470           2489678 ns/op
BenchmarkPoseidonLarge-4             476           2530568 ns/op
```

- With the optimizations of #12:
```
BenchmarkPoseidon-4                  766           1550013 ns/op
BenchmarkPoseidonLarge-4             782           1547572 ns/op
```

- With the changes of this PR, where uses goff generated code instead of *big.Int:
```
BenchmarkPoseidon-4                 9638            121651 ns/op
BenchmarkPoseidonLarge-4            9781            119921 ns/op
```