ci: fix failed runs

[netrunnereve]

This fixes the broken #26593 thread test by moving it to a Github machine, fixes the occasionally failing cpu x86 run, and makes the cmake run runnable on all machines (tclsh comes with Ubuntu and I think all our self hosted Linux machines are Ubuntu).

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: NO

[ggerganov]

Let's try, but there is a risk that the builds will be very slow on the Github runners without a ccache (and we cannot afford to use ccache because of too much space).

[CISC]

Let's try, but there is a risk that the builds will be very slow on the Github runners without a ccache (and we cannot afford to use ccache because of too much space).

As long as it's below 2hrs it's still a win. :)

[ggerganov]

Let's try, but there is a risk that the builds will be very slow on the Github runners without a ccache (and we cannot afford to use ccache because of too much space).

As long as it's below 2hrs it's still a win. :)

It's a tricky balance. The Github runners are limited (20 in total). If the jobs are slow, they occupy a slot and block it for a long time. In such cases, it's better to keep running them on the self-hosted runners where they complete very quickly (mainly thanks to the ccache) - even if it means staying in the "self-hosted" queue a bit longer.

[ggerganov]

Let's wait to see how long the THREAD and ADDRESS runs take here before merging. Will take a bit since currently the Github runners are working on the new llama.app release: https://github.com/ggml-org/llama-install.sh/actions/runs/31591724765

[netrunnereve]

Let's try, but there is a risk that the builds will be very slow on the Github runners without a ccache (and we cannot afford to use ccache because of too much space).

In this case it's probably worth tossing the Vulkan SDK cache, it's 500mb but I don't think it takes that much time to install and download. We can also look at the list of jobs again and see if anything can be cut down more.