feat(windows): preserve running nested vm execution state

[toksdotdev]

a running nested vm can fail after intel windows restore because whp reloads stale ordinary vmcs backing memory. enable the documented enlightened-vmcs interface on capable intel hosts so linux keeps that state in guest memory, and reject captures that still use an ordinary active vmcs.

save the full emulated pit timer state alongside cpu and interrupt state. without it, older ordinary snapshots can restore with a frozen guest clock and hanging commands. pause fences timer injection, and a high-resolution windows timer preserves millisecond deadlines without breaking amd startup.

ordinary snapshots now use whp abi 2 and enlightened nested snapshots use abi 4. normal restore rejects older snapshots with recovery instructions. ExecutionState::recover_windows_periodic_pit(divisor) explicitly supplies a known periodic reload for old ordinary abi 1 or enlightened nested abi 3 snapshots; the original timer phase cannot be reconstructed. validate the recovered guest and capture a new snapshot. non-enlightened nested snapshots still require a disk restart. cpu compatibility checks remain enforced.

preserve the hyper-v clock epoch and synthetic feature contract, and restore nested state, registers, xsave and interrupt state together. amd keeps its existing guest features.

merge order: #156 → #155 → #158 → #159 → #160 → #162 → #169 → #168. windows arm64 and linux arm64 remain unsupported; #164 is deferred.

Confidence Score: 5/5

The reviewed changes appear safe to merge, with no new blocking issues found.

_{Reviews (10) · Last reviewed commit: "fix(windows): use precise interruptible ..." · Reviewed by Greptile}