cvm: Refactor ohcl/ovmm emulator for flexible register retrieval and update

[babayet2]

note: still a couple active discussions about this PR, but no longer in draft mode so we can see CI results

Motivation

1. When an instruction is emulated for a CVM guest, the full guest register state is retrieved. For TDX, this dispatches an expensive TDCALL to read each segment register regardless of whether it is necessary for the emulation.
2. The existing emulator code allowed arbitrary guest registers to be written: if the register is considered immutable, the update is silently dropped.

Changes

• The emulator frontend no longer has a guest state struct. Instead, the frontend calls functions to read and update register state.
• Each backend defines and implements its own register state struct. On a register read/write, the backend has the flexibility to choose whether a register should be read from this struct, or pass through directly to the backing. With the exception of segment registers for the TDX backing, the existing behavior has been preserved, e.g. reading XMM registers passes through to the backing, and GPs are read from the state struct.

[smmalis37]

Take a look at #478 too, lets make sure we don't conflict too hard.

[smalis-msft]

We should add some documentation on which registers get set immediately, and which only get set when you call flush.

[smalis-msft]

And which are immutable i guess.

[smalis-msft]

I guess these docs would have to be per-backing.

[smalis-msft]

Overall this looks really nice, and a great step in the right direction. Thanks!

[babayet2]

@smalis-msft taking this out of draft mode as several of the CI tests don't fail to run locally for me, I'd like to see the output of the full run

[smalis-msft]

This looks pretty good to me overall. I've asked @jstarks to also take a pass on it as well though. Also note that all of our windows-hosted vmm tests in this repo rely on virt_whp support, so they'll probably hit your todos. SNP/TDX/MSHV testing is done in an outer loop currently. (We know that situation isn't great and are actively working on making it better).

[smalis-msft]

This needs to be a flush now right? Or is that handled inside emulate_mnf_write_fast_path?