Read the Removed bitmap from PageTracker and emit it as DiffMetadata.Empty
so REMOVE'd pages become uuid.Nil mappings in the snapshot header (read as
zero on resume). Defensively AndNot the empty set out of dirty: settle drains
make these disjoint in practice (Removed pages have no PTE, WP-async only
sees present pages with WP cleared), but if the invariant ever breaks the
guest's last intent for a Removed page is "free, read zero on restore" — so
empty must win, not stale dirty content.

…e comments

Receiver derives freePageReporting from FC version + LD flag and never
reads cfg.GetFreePageReporting(), so the proto field was dead. Removing
it also kills the noisy generated pb.go diff.

A REMOVE'd page that gets a read fault before snapshot installs
zero+WP+wake. The unconditional Zero->Dirty promotion stripped it from
the empty bitmap while WP-async dirty scan still excludes it (WP set,
no write yet), so the page fell back to the parent layer's stale
content. Keep the tracker entry as Zero for that case and let WP-async
surface any later write; flip the AndNot so dirty wins over empty for
written pages.

…e-reporting-integration

Remove the !hugePages guard — hugepages and balloon FPR are not
mutually exclusive in Firecracker.

Generic name and parameterized FreePageReporting so individual balloon
features (FPR today, FPH next) can be opted in independently from the
caller without renaming the helper again.