Functions to generate random integers

[ghost]

This pull request is for functions generating random integers.

randomint/0 returns an unpredictable value between 0 and 0xffffffff (included). randomint/1 function returns an unpredictable value between 0 and given upper bound (excluded) randomint/2 function returns an unpredictable value between lower and upper bounds (included)

NOTE: The documentation for randomint/2 may be incorrect.

[jlouis]

Sorry for letting this sit for too long.

Some comments:

• I've just pulled an implementation of the pwhash functions (and provided EQC tests for these), so we can probably strip those out of the PR.
• If I read the commits correctly, it seems like the new stuff is from 8857bf7. This suggest we should rebase the PR and only cherry-pick these new commits. I think everything else in the history is accounted for. It also means we can rid ourselves of merge-commits
• We should consider calling randomint() for random_int32() to make it explicit what it actually returns. I think Erlang readers might try to write stuff like randomint(0, 10000000000000000000000) and expect this to work.
• is this different from <<I:32/integer>> = enacl;randombytes(4) in any way? I know the /1 and /2 variants are, but the /0 variant is what I'm thinking about. In short, do we really need this interface?
• Supporting a true integer interface with arbitrary sized integers would be nice, but we can start with the 32bit variants first I think.

[jlouis]

Long overdue, but I've implemented enacl:randombytes_uint32/0 and enacl:randombytes_uniform/1 on master now.

There are some changes compared to this proposal:

• The name makes it more obvious you are getting a 32bit (unsigned) integer out of this. It should avoid problems where people think they are getting something, which they are not.
• The uniform variant fails on integers above uint32 MAX. I'd rather crash peoples code.

I've opted to not implement randombytes_uniform/2:

• It isn't in libsodium. It isn't in e.g., the Go rand package.
• It is fairly easy to implement yourself: Just add the lower bound, and use N+1 if you want the upper bound included.