ClusterFuzz shouldn't exception on new OSS-Fuzz projects that fail to build for the first time.

Reverting failed first-time builds is a frequent sheriff duty. I think we could avoid it if ClusterFuzz only created jobs for projects that have at least one build. CI has reduced the need for this a lot. But it isn't foolproof: https://github.com/google/oss-fuzz/pull/4393 CI fails to reflect our actual builds from time to time (in this case the disk, which is very small on GH actions, filled up). Fixing this would help make OSS-Fuzz more self-service.
I frankly think it's probably a good idea to never give resources to jobs where the last build failed. It's good incentive to fix things for projects who's integrations are actually maintained and it basically eliminates the need to prune the large number of projects that dont fix builds for months.
A less thorough fix would be using GCB for CI so CI more accurately reflects whether the build will fail.
It mitigates this problem but doesn't fix it though since upstream can break something in between CI working and the first build happening.