mem_space: support accesses with an offset

[Qwinci]

Add versions of read/write that take in an additional offset. It is useful for accessing register arrays, previously you had to do something like

auto value = arch::scalar_load_relaxed<arch::bit_value<uint32_t>>(space, regs::myRegStart + index * 4);
value |= myReg::value(true);
arch::scalar_store_relaxed(space, regArrayStartOffset + index * 4, value);

whereas with this change you can do

auto value = space.load_relaxed(regs::myReg, index * 4);
value |= myReg::value(true);
space.store_relaxed(regs::myReg, index * 4, value);

[avdgrinten]

There are some alternatives to this that are worth considering:

• You can use a function that returns a register as we do in some places (e.g., https://github.com/managarm/managarm/blob/master/kernel/thor/arch/riscv/plic.cpp#L19):

  // Per interrupt-source registers.
  constexpr auto plicPriorityRegister(size_t n) {
  	return arch::scalar_register<uint32_t>{static_cast<ptrdiff_t>(4 * n)};
  }

• You can use subspace() to do something like space.subspace(offset).load_relaxed(some_register).

[avdgrinten]

Is still useful to have functions that take an offset despite these alternatives being available? I'm not necessarily against the new functions but I'd like some justification why the existing approaches are inconvenient.

[Qwinci]

I guess the approach with a function is pretty nice, I didn't think of that. The functions that take in offsets aren't probably needed over that though I think they could possibly still be a nice option for e.g. registers where the offset depends on multiple runtime variables (as in that case for the function approach you'd have to either pass multiple parameters/have it as a member function and the subspace method is kinda long).