This is all asuming you don't have a server class gpu
1) What physical resource does this come from? The bare metal host's RAM? The host's dedicated video memory? I'm aware there is the ability to passthrough a GPU, but this pertains to a bare metal host running only onboard video card.
The normal hosts memory
2) How can I monitor a VM's video memory usage? i.e. how do I prove out that the VM's VRAM is at capacity and needs more, like a dedicated GPU card? GPU isn't show in task manager as it is on physical machine.
You can't its part of the VM processes
2b.) How can I see what the overall VRAM usage is, if it's at capacity or not?
In-guest tools like Liquidware labs may be able to do this, but I've only seen them do this with server gpus
Overall my question is if you need to check this for anything other than an intellectual exercise, you probably should have a server GPU. In general most servers don't need gpus to do there normal functions, the memory used to display a console or the windows interface doesn't use a lot and I've never seen an issue. Desktops are different since they build in gpu acceleration since they assume there is some sort of gpu included, even an onboard one like intel