add validation for nestedVirtualization, rosetta, and mountType

[olamilekan000]

change validates config values for `nestedVirtualization`, `rosetta`, and `mountType` 
when a user starts or tries to create a VM.

Fixes #3126

[olamilekan000]

@AkihiroSuda kindly review. Also, I am not sure why unit tests are failing for windows, go version 1.22 and 1.23, even though everything looks fine.

[AkihiroSuda]

https://github.com/lima-vm/lima/actions/runs/12853756300/job/35837296730?pr=3127

=== RUN   TestValidateRosetta
time="2025-01-19T12:59:05Z" level=debug msg="ResolveVMType: resolved VMType \"qemu\" (explicitly specified in []*LimaYAML{o,y,d}[1])"
    validate_test.go:202: assertion failed: expected an error, got nil
--- FAIL: TestValidateRosetta (0.00s)

=== RUN   TestValidateMountTypeOS
time="2025-01-19T12:59:05Z" level=debug msg="ResolveVMType: resolved VMType \"qemu\" (default for GOOS=\"linux\")"
time="2025-01-19T12:59:05Z" level=debug msg="ResolveVMType: resolved VMType \"qemu\" (default for GOOS=\"linux\")"
time="2025-01-19T12:59:05Z" level=warning msg="`mountType: virtiofs` on Linux is experimental"
    validate_test.go:272: assertion failed: expected an error, got nil
--- FAIL: TestValidateMountTypeOS (0.01s)

[olamilekan000]

I have tried to replicate it locally but it passes. Not sure what I'm missing

[olamilekan000]

@AkihiroSuda Found and fixed the issue. Thanks

[jandubois]

@olamilekan000 I don't know if all your pushes where necessary to deal with CI problems, but please be aware that each full CI run costs about $6 due to our use of large GitHub runners for macOS, and this PR had like 18 force-pushes¹. That's ok if the problem is only reproducible in CI, but otherwise please try to resolve problems locally first.

Footnotes

1. I have not checked, maybe the macOS jobs where skipped due to earlier compile/lint failures, in which case the cost will have been minimal. ↩

[jandubois]

I've only reviewed the validate changes, not the tests.

There may be more conditions that the validation could check for, but I'm out of time for now to think about it more.

[olamilekan000]

@olamilekan000 I don't know if all your pushes where necessary to deal with CI problems, but please be aware that each full CI run costs about $6 due to our use of large GitHub runners for macOS, and this PR had like 18 force-pushes1. That's ok if the problem is only reproducible in CI, but otherwise please try to resolve problems locally first.

Footnotes

1. I have not checked, maybe the macOS jobs where skipped due to earlier compile/lint failures, in which case the cost will have been minimal. ↩

Thanks for bringing this to my attention. The reason for that many pushes was due to me trying to figure why the test passes locally but fails with the CI. Apologies for the inconvenience

[jandubois]

There are still invalid setting combinations that are not covered by tests.

It is a bit unfortunate that complete test coverage can only be achieved by running the test on each supported os and arch.

I think it would be better if the validation code had global variables for os and arch that can then be overridden in the tests to check for all scenarios with just a single local test run. But that should be a separate PR.