Unless we are fuzzing `CheckProofOfWork` itself, it makes little sense
to have it be a blocker when fuzzing. We therefore use
`FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION` (only defined for fuzz builds
by e.g. clang with -fsanitize=fuzzer, afl-clang-lto, etc.) to replace
the actual proof-of-work checks with checking if the first bit on
`nNonce` is set. This allows the fuzzer to create PoW (in)valid blocks
by controling that one bit instead of having to grind for actual PoW.

Checking for a bit in `nNonce` is preferable over checking for a bit in
the block hash as `nNonce` is chosen by the fuzzer and when whole chains
of blocks are created during fuzzing a change in an old block won't
invalidate the fake PoW on newer blocks.