lib: add better management of Hyper-V overlay pages

[gjcolombo]

Add to the Hyper-V module some supporting types to support overlay pages. A Hyper-V overlay page is a sort of auxiliary page of guest memory that "covers up" or temporarily "replaces" a specific physical page in the guest's physical memory, such that accesses to the overlaid GPA end up accessing the overlay instead. See the doc comments throughout overlay.rs for many more details.

To avoid creating a 2,500-line patch, this PR omits direct migration of overlay manager state. Instead, the target Hyper-V stack re-creates all overlays during import based on the contents of its imported MSRs. This is somewhat dangerous: a single guest page is allowed to have multiple overlays, and there's nothing in this change that guarantees that the source and target will agree on which overlay is active at migration time. The danger is mitigated here by having only one overlay page in the whole VM (the hypercall page), but this will need to be addressed in a follow-up PR before adding any more overlay kinds (like a reference TSC page).

Tests: cargo tests, including new unit tests of the overlay manager; PHD runs with Alpine and Debian 11 guests; manually migrated a VM and made sure the hypercall page was preserved properly over migration.

Related to #850 (but doesn't close it since migration work is still pending).

[gjcolombo]

Thanks for the review, @hawkw! I think this is ready for another lap at your convenience. I'll work on rebasing #853 onto this.

[hawkw]

The new changes look good to me! Perhaps worth getting a second opinion?

[gjcolombo]

I found a bug during ad hoc testing: HyperV::reset creates a new overlay manager, but doesn't attach its (new) memory accessor to the accessor hierarchy. It needs to do that (and reset needs to be careful to drop things in the correct order; it does so today, but only by the grace of the drop ordering rules for structs...).

I'm going to merge #849 and rebase this branch before pushing a fix and working on other PR feedback.

[gjcolombo]

I found a bug during ad hoc testing

a3186ef should fix this.