Consider accepting interrupts while in VMX root context

[ALSchwalm]

Currently we always mask external interrupts in the hypervisor context. This simplifies the design, but can hurt latency characteristics for high priority interrupts. Consider the following situation with our current logic:

1. Interrupts are not masked in the guest
2. An external low-priority interrupt arrives triggering a vmexit
3. As the vmexit occurs, external interrupts are masked
4. While processing the interrupt, a higher priority interrupt arrives
5. Because interrupts are masked, nothing happens with this interrupt immediately
6. After finishing processing the interrupt, it is injected in to the guest and the guest is re-entered
7. At this point we should get a vmexit for the high-priority interrupt
8. We process and ultimately inject the high-priority interrupt
9. Now the guest jumps to the high-priority handler

Having interrupt handling at the hypervisor level would essentially allow us to coalesce interrupt delivery in a single vmexit, which should improve latency properties. However, this would be a significant change to the hypervisor design and should not be done without a lot of planning.