Fix unaligned cast in impl_uint_from_fill

[pitdicker]

It seems to me doing an unsafe cast from [u8; 4] to u32 is not right, because they have different minimum alignments. But this code was used by both OsRng and ReadRng, and did not seems to cause problems. No idea why.

[dhardy]

I would guess the byte array was sufficiently aligned anyway.

Does this change affect benchmarks?

[pitdicker]

I think LLVM optimizes out a memcopy here and writes fill_bytes directly to the u32 in the parent stack frame. So in practice it is was already aligned well. So no real problem here, but it seems neater to me :-).

I will run a benchmark, but even if it compiles to different instructions I doubt it is something we can measure.

[pitdicker]

There is no measurable change in benchmarks.