Split functions between architectures, make register access type-safe

[funsafemath]

Registers are now correct by construction, so reads/writes now return T instead of Result<T, E> (hopefully unicorn never returns an error for a valid register), also CPU-specific associated functions like reg_write_arm_coproc are now implemented parametrized types like Unicorn<D, Arm>

The arch parameter should probably be the first one, as the data parameter is usually inferred or is equal to the unit type, and it may be a good idea to set a default value for it (unit type), which is possible only if it's the trailing generic parameter