net: fix missing network address when emulating a foreign arch

[anandghegde]

Fixes #1304
Fixes #1556

Problem

On Apple Silicon, colima start --arch x86_64 --network-address starts fine but the VM never gets a reachable address: colima ls has an empty ADDRESS column and colima ls -j has no address field. The same happens with --vm-type vz --vz-rosetta --arch x86_64 (#1556). @djzort's comment in #1336 shows the pattern clearly: with the default vm type there is no "preparing network" step and no address, while adding -t qemu prompts for the sudo password, prepares the network and the address shows up.

Cause

vz and krunkit can't run a foreign architecture, so newConf quietly falls back to QEMU when the guest arch differs from the host. QEMU needs vmnet to attach col0.

startDaemon, however, decides whether vmnet is needed from the configured vmType. With the default vz it concludes vmnet isn't needed, drops Network.Address, and never starts the vmnet daemon. newConf then takes the QEMU/vmnet branch, finds vmnet not enabled in the context, and skips col0 without a warning.

Fix

Pull the "which VM type will Lima actually use" logic out of newConf into limaVMType(conf), and use it from both newConf and the vmnet decision in startDaemon (usesVmnet(conf)). There is no behaviour change for same-arch vz/krunkit/qemu setups.

Testing

• New Test_usesVmnet and Test_limaVMType in environment/vm/lima. They are host-independent: the "foreign" arch is whichever one the runner isn't.
• Test_usesVmnet/vz_foreign_arch_falls_back_to_qemu fails against the old condition (usesVmnet() = false, want true) and passes with the fix.
• go test ./..., go vet, GOOS=linux go vet ./environment/vm/lima/ and golangci-lint run (0 issues) all pass on macOS 26 / arm64.
• I haven't done a full x86_64 boot with the fix: that needs qemu plus sudo for the socket_vmnet install, which I can't do on this machine. The colima list shows no IP address #1336 comment above is the field confirmation that the QEMU + vmnet path yields an address.

Side note, not changed here: colima status still says "macOS Virtualization.Framework" for these instances because DriverLabel also reads the configured vmType. I can follow up if you'd like that fixed too.

AI disclosure: implemented with Claude Code (Claude Opus 5). I reviewed the change and ran the tests above.

[abiosoft]

Only human commits are allowed.
Also, kindly sign your commit(s).

[anandghegde]

@abiosoft Commit is not signed

[abiosoft]

Kindly remove claude as a co-author of your commit. Only human committers are allowed.

[anandghegde]

Done — dropped the co-author trailer and force-pushed. Commit is otherwise unchanged and still signed 👍

[abiosoft]

Lastly, can you kindly rebase, just to ascertain there would be no conflicts.

[anandghegde]

Rebased onto latest main (f450fa7) — no conflicts, build and go test ./... pass, and the commit is still signed.