jets: fix buffer overrun

[dozreg-toplud]

Wasn't that a buffer overrun? buf_w[8] is outside of the atom buffer.

[joemfb]

We discussed this in ccrs. This fix is correct, the old code was wrong, but almost always (1 / (2^32 - 1)) effectively correct anyway (since random garbage will likely not have all bits set, and a refcount will certainly not have all bits set).

[dozreg-toplud]

I think this reasoning applies for well-formed input, i.e. input for which this function should return one. I guess if a jet was called in such a way that this function got an eight word-long atom which was greater than now_w buffer we would get a false positive and a jet mismatch: the function would compute some result instead of crashing.

If in practice that function was called with only well-formed inputs then there's nothing to worry about.