fixes for macos CI

[bashbaug]

fixes #144

This PR includes two changes to fix macos CI:

1. Set the SDKROOT environment variable so PoCL can find system libraries, see discussion on the issue.
2. Disables the callbackcpp sample from testing, since it is producing segfaults for release builds only. I'm not sure if this is a PoCL issue or a sample issue, but it will need additional debug.

[bashbaug]

Sigh... of course now there's a completely new error I didn't see in any of my testing.

I'll need to look at this again.

[tneval]

@bashbaug how is it that this wasn't caught in
https://github.com/bashbaug/OpenCL-SDK/actions/runs/17270551550/job/49013861064 ?

[bashbaug]

how is it that this wasn't caught

Yeah, it's frustrating to me and it doesn't make any sense 😄

I was hacking a bit on the CI files in my branch while testing, but it should have only been to enable testing on my branch and to disable the configs I wasn't modifying. I don't think I did anything that would have caused this failure to go away, or anything in this PR that would cause it to appear.

Note that the "simple readme build" passed on macos for this PR, so it has something to do with the more complicated testing. Hopefully I'll get more time to look at it in the next couple of days.

[tneval]

how is it that this wasn't caught

Yeah, it's frustrating to me and it doesn't make any sense 😄

I was hacking a bit on the CI files in my branch while testing, but it should have only been to enable testing on my branch and to disable the configs I wasn't modifying. I don't think I did anything that would have caused this failure to go away, or anything in this PR that would cause it to appear.

Note that the "simple readme build" passed on macos for this PR, so it has something to do with the more complicated testing. Hopefully I'll get more time to look at it in the next couple of days.

Heh, I see. Was just thinking that it's possible that the env var 'hack' doesn't work with all MacOS/Xcode versions.

[bashbaug]

OK, it looks like it's all building and passing if I move the PoCL set up to its own step. This could just be coincidence, but I'll take it! I'll mark this PR as "ready for review" again.