The Synthetic PoRep code is interleaved with the original PoRep code. This commit
refactors the code, so that it's clearer what the differences between Synthetic
and Non-Synthetic PoRep is, without the need to read through large parts of the
code, looking for if-statements.

It also becomes more apparent in the code, that there are two "modes" of the
`prove_layers` function in case of the Synthetic PoRep. One generation step in
case there's no seed and one extracting step when there's a seed.

There no longer is a fallback in case from Synthetic PoRep to Non-Synthetic PoRep
in case the Synthetic PoRep Proofs file cannot be read, but when the labels are
set. I consider that a feature as if this case happens, your system likelt has
some problems, so it's better to error early.

`prove_layers_generate()` now operates on a single partition. This makes it
clearer that Synthetic PoReps always operate on a single partition only.

With the challenge generation moved outside of this function, it's also a
clearer separation of concerns, as it now operates directly on the challenges
given, without know anything about how they were generated. This also useful
for the work of having small, special case binaries for certain operations.