[clang] [llvm] [ci] New script to generate test reports as Buildkite Annotations (PR #113447)

[David Spickett via cfe-commits]

DavidSpickett wrote:

> Do we make any guarantees about supporting lit flags in the future? It's mostly an internal tool, and I would personally be in support of removing flags that are no longer used in tree, unless there are some extremely compelling downstream use cases and there is an appropriate maintenance story.

> Overall, I'm fine with this patch, however I wonder how much effort it is worth putting into the BuildKite infrastructure given that we plan on moving as much as possible to Github Actions...

Louis' point is a good one given that lit is released externally (https://pypi.org/project/lit/). We probably don't have a contract as such there but if we don't add anything, there's no chance to annoy anyone when we remove it. The next release would be based on the 19 branch, so there is still time to remove the feature I added.

The alternative is to go back to changing the build scripts to do:
```
for target in targets:
   ninja target
   move result file
   accumulate return code
```
That does not require a lit feature but is more invasive. If we're going to throw away those scripts anyway, perhaps folks will be ok with a temporary increase in complexity.

I agree that a future GitHub actions pipeline should be splitting check targets into steps anyway, effectively doing the for loop I showed above, but using the pipeline structure. And if the folks doing the transition don't want to worry about that on top of everything else I'm sure they have to do, I will commit my time to doing the extra work required for that.

If that means we get useful reporting of test results by default and we can delete every change I've made towards this so far, I'll be very happy!

I will make another version of this PR that modifies the build scripts instead of using the lit feature. If that PR is preferable, I will revert the lit feature.

https://github.com/llvm/llvm-project/pull/113447