SMP fixup - WIP

[CharlyCst]

WIP: do not act on those items yet, the list is evolving (and might be corrected). Plus we will need to prioritize properly.

This issue is a collections of things that will eventually need to be fixed for SMP to properly work on real hardware and to ensure we can reliably debug and develop the hypervisor going forward.

• Fix this hardcoded value: https://github.com/CharlyCst/vmxvmm/blob/5b3bf8628841e329e3e6b2d866fbd6e3fcd03a58/vmm/first-stage/src/acpi_handler.rs#L15
  This might not work on a physical machine...
  I assume it is the virtual to physical offset, in which case we can either pass it to the struct or find a better way of handling it across the whole project (e.g. using a global value, considering it is initialized only once).
  - Same as above in the apic implementation? (missing a file currently, but I assume moved from this file: https://github.com/CharlyCst/vmxvmm/blob/c9bd039c5c5b1212305a61c1071f63f6cc0f179a/vmm/first-stage/src/interrupts.rs#L40)
• The double fault stack is shared among all the cores (because it is a global static):
  https://github.com/CharlyCst/vmxvmm/blob/c3fa691cd70a4d4d6775e4dae5dbb04d532ca63d/vmm/first-stage/src/gdt.rs#L32
  This is not that big of a deal because it will only cause bugs if we already have bugs (i.e. a double fault), but sill, better to be aware of it.
• I'm not sure why print_vmx_info has any side effect, it performs read-only operations according to my understanding. Investigate why this poses issued in the "smp: setup vmxon environment on APs" commit (e01f565)
• We should serialize all accesses to the data used by the hypercall layer by simply adding a lock on hypercall itself, and not creating per-core hypercall instance.
  The implementation proposed, such as https://github.com/CharlyCst/vmxvmm/blob/c527ca71d5fbd0b5019c9700c270e9a69dc9e73b/vmm/second-stage/src/x86_64/guest.rs#L74
  Is unsound as it can return multiple mutable references to the same object.