Refactor state PDA accessors

[fedgiac]

First step for introducing the management of solver accounts. Nothing change in the code behavior yet, it's just a refactor in preparation for new code.

Right now, the handling of the state PDA is awkward: we have an EncodedStateAccount that takes ownership of the data and a lot of accessors like EncodedStateAccount::authority(bytes, role) that take a bytes array to edit in place.

With these changes, we drop EncodedStateAccount and replace it with StateAccount. Notably, StateAccount doesn't own its data but it takes a reference. This allows us to access the data with things like state.authority(role), which reads very naturally and allows us to drop the bytes accessor and gives us a type-safe way to store the state account bytes.

The nicest trick in these changes is making the implementation generic in the Deref'ed type:

impl<T: Deref<Target = [u8]>> StateAccount<T> {...}
impl<T: DerefMut<Target = [u8]>> StateAccount<T> {...}

This means that if we use StateAccount(&bytes) then we only access read-only methods, while StateAccount(&mut bytes) also gives access to mutable functions on the ref. The code duplication between & and &mut is something we struggled with in the past and is imho nicely resolved by this code. All write and read accesses are, naturally, zero copy. This will work nicely for the code that updates solvers.

Note also that the state account now has a header. In this PR, the header is the state account. Once we add solvers, they will not be part of the header, and we can keep most of the remaining code unchanged. The layout doesn't change, yet, though it theoretically now supports extra padding bytes (ignored).

Role::ALL

I added Role::ALL because it was helpful in a test. It isn't used outside of tests but I still decided to make it part of the interface since it's properly tested and can be helpful. I also used it in other parts of the code where it would be helpful.

Tests

Updated tests work.

[kaze-cow]

I love this change. tbh this is kind of the direction I was hoping for before!

based on what I have seen here, I would be in favor of later refactoring the order PDA similarly, as we can still benefit from zero copy reads there.

[kaze-cow]

only one followup thing to review, otherwise all addressed.