Write an SP1 abstraction layer

[Ekleog-Polygon]

With an SP1 abstraction layer such that we only ever use SP1 through it, we'll become able to kill the SP1 wrappers from #263, and instead just handle panics from within the abstraction layer.

In particular, it should also wrap (de)serialization so that any panics from (de)serializing SP1 structs get catched without needing any wrappers.

Once this gets done, we will be able to more easily consider any multi-prover project that could arise in the future, and we will no longer need to sprinkle catch_unwinds around our code.

[atanmarko]

First iteration: define the shape of the system of how we use proving system dependencies and what would be the best way to abstract it.