Should the accelerator fallback to "tcg", if the default accelerator is not available?

For instance if /dev/kvm is missing on Linux, possibly because running in a VM

Is this really necessary? I know we say it is for "debugging", but you can already override QEMU settings with the QEMU_SYSTEM_X86_64 and QEMU_SYSTEM_AARCH64 environment variables. It is a bit clunky to use, but is this really a feature for more than a single user?

Having a whole matrix of accel types feels very overengineered for lima.yaml.

Same as with CPU types, this feels more like something that should go into a local config file, and not into the individual templates. Like we have $LIMA_HOME/_config/networks.yaml.

I understand that with this PR the setup can go into a $LIMA_HOME/_config/default.yaml, but I wonder if it is right to make the regular template schema more and more complex with settings that are not really instance-specific.

I think we always add the -accel flag even if kvm is missing, but I guess you can filter it out in a custom script wrapper

You can do this:

> export QEMU_SYSTEM_AARCH64="/opt/homebrew/bin/qemu-system-aarch64 -machine virt"
❯ limactl start --tty=false --vm-type qemu
…
WARN[0001] [hostagent] Specifying args ([-machine virt]) via $QEMU_SYSTEM_AARCH64 is supported only for debugging!
WARN[0002] [hostagent] Not adding QEMU argument "-machine" "virt,accel=hvf", as it conflicts with "-machine" "virt"
…

It then fails for reasons I don't have time to investigate right now, but the basic mechanism works: if you specify -machine in the environment, then it won't be added by Lima.