Setup fuzzing with afl

[slaperche-scality]

afl is a kind of "intelligent" fuzzer that feeds "random-not-so-random" inputs to a software to test its robustness.
It uses compile time instrumentation to monitor in real time which code path is covered by the input, and then works by mutating the input to try to reach new uncovered code path.
When it trigger a crash, it saves the input (so we can replay it to debug).
It's kind of compiler-guided fuzzing, which is more efficient than random fuzzing and cover more code than manual fuzzing.

Codec are a very good target for this (since they are supposed to accept almost anything in input without crashing).

As fuzzing is a long-running process (i.e not sth that can be done in the CI) we could imagine having a VM that run continuously a fuzzer on some specific tag (like a beta of our next release ot sth like that).