CI: Bring back ccache for DPC++

[psalz]

I just randomly noticed that CI run times are going absolutely through the roof again, with runs taking anywhere from FOUR to SIX hours. While not really addressing the root problem, our fix for this so far was to use ccache, which has worked rather well.

It looks like ccache was lost in the switch to the DPC++ nightly containers (#712), so with this I'm bringing it back.

Note: I haven't tested this, so after the cache has been primed by the CI run for this PR we'll have to launch a second run to see if it actually works.

---

Side node / reminder: we generate a detailed report of what is taking up the time, looks like the marray operators test is the new champion with a whopping 25 minutes - in case anybody from Intel wants to take a look.

[keryell]

Looks good.
Do you remember why we do not use https://github.com/hendrikmuhs/ccache-action like in https://github.com/triSYCL/triSYCL/blob/master/.github/workflows/cmake.yml#L123 ?

[bader]

@psalz, unfortunately, it seems like it doesn't help.

I still see that cache is not used here: https://github.com/KhronosGroup/SYCL-CTS/actions/runs/5570325262/jobs/10174289342. This is for the #740, where I merged this change.

BTW, I think ccache is installed in the DPC++ image by this script: https://github.com/intel/llvm/blob/sycl/devops/scripts/install_build_tools.sh#L7. We run it to build "base" image, which we use to build the container with "HW drivers". The nightly build is based on the "HW drivers" image.

[psalz]

Yes the merge was a bit premature (I did say I'm not sure whether it's working or not). I'm debugging it now.