Add simple fuzzing harness for functions with floating-point parameters (such as ser_double_to_uint64(double), etc.). Add serialization/deserialization fuzzing for integral types.
Backport of core PR17996.
The first and last commit are not relevant to us.