ci: build armv7-gnueabihf with canonical runner

[ubiratansoares]

Canonical has made more runners available for us, in particular these armv7-linux-gnueabihf ones.

Creating this PR to give them a try and validate the integration. Ignore this PR for now.

[tgross35]

I'm excited for this one!

[tgross35]

What makes these armv7 runners? https://github.com/rust-lang/compiler-builtins/actions/runs/35132215572/job/104915778763?pr=1320#step:3:21 and https://github.com/rust-lang/compiler-builtins/actions/runs/35132215572/job/104915778763?pr=1320#step:6:168 looks like it's an aarch64 host

[ubiratansoares]

What makes these armv7 runners?

I was told it and I took it for granted, but let me report this back to Canonical just in case

[ubiratansoares]

So, according to what I got from Canonical

the runner itself is a genuine 32-bit armhf build. the ci log ( uname -a = aarch64 ) is just the host kernel reporting its own native architecture as these VMs run an arm64 kernel with AArch32/EL0 support enabled, which lets 32-bit armhf binaries execute natively

In this case, I think we can keep the arm64 host toolchain, cross-link with the armhf gcc, and run the test binaries natively in AArch32 mode, ie with no qemu. That would require adding a few additional setup steps via ci/install-test-deps.sh if I understand the ci harness correctly.

@tgross35 does that make sense to you?

[tgross35]

Makes sense, I’m just surprised that it seems like arm32 isn’t reported anywhere and that rustup decided on an aarch64 host.