The dispatcher image makes reproducibility hard

Sometimes old experiments can't be reproduced easily without manually modifying FuzzBench to use the dispatcher-image from the original experiment. This may be true for the Q based measurer when that is implemented.
To remedy this, I have two changes in mind:
1. Experiments should save data on which dispatcher image they used and this info should be automatically used in reproduction experiments.
2. the clang coverage binaries should be included in the coverage builds instead of on the dispatcher image. This is how OSS-Fuzz does it and it makes most sense.