[sanitizers] Improve size of blame lists of slow builders

[vitalybuka]

According to llvm/llvm-project#115361 (comment) unrelated steps of current bootstrap bots unnecessarily spam libc++ patches.

[vitalybuka]

This will not be helpfull.
The reason is that those bots are quite slow 3-4 hours.
If libc++ patch gets into blame list with another bad patch, it will be notified, regardless we run or not check-cxx.

The problem is make them faster in general.
They run on quite fast hardware.

Without dropping test coverage, or significantly increasing number of VMs, I don't see good way to have blame list small.

[asb]

This will not be helpfull. The reason is that those bots are quite slow 3-4 hours. If libc++ patch gets into blame list with another bad patch, it will be notified, regardless we run or not check-cxx.

The problem is make them faster in general. They run on quite fast hardware.

Without dropping test coverage, or significantly increasing number of VMs, I don't see good way to have blame list small.

I think the solution is the "gatekeeper" idea that @gkistanova as mentioned before and I think had a prototype implementation at one point. Basically - only kick off a build on a slower builder once a faster builder has completed successfully. This reduces the chances of creating a delayed failure response for an issue that was already found and fixed through bots with a faster turnaround time.