Incorrect discovery callback handling leads to sporadic tests failures

We are starting to see a lot more pipeline failures, especially for win64. This seems to be due to an incorrect handling of the discovery progress, especially when there are more than a single entry point being queued.

Most likely this happened when we split discovery tasks in 2 (!386 (merged)) which causes the following sequence to occur:

discover( entrypoint 1 )
onEntryPointAdded( entrypoint 1 )
onDiscoveryStarted( entrypoint 1 )
discover( entrypoint 2 )
onDiscoveryCompleted( entrypoint 1 ) // Since the first discovery is interrupted to add the 2nd entrypoint
onEntryPointAdded( entrypoint 2 )
onDiscoveryStarted( entrypoint 1 )
onDiscoveryCompleted( entrypoint 1 ) // Now it actually finished, but since we expected only 2 discoveries, we consider the discovery to be fully completed, which is false
onDiscoveryStarted( entrypoint 2 )
onDiscoveryCompleted( entrypoint 2 )

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information