Speed up orbit computations on integer sets and tuples by providing a better hash function for lists of small integers

[reiniscirpons]

This PR improves the hash function used for lists.

Text for release notes

see title

Further details

The existing function essentially determined the hash by onlyconsidering the first two entries of the list. This causes significant lots of hash collision and slowdown when computing the orbit on sets, e.g. before the changes in this PR we get:

gap> Orbit(SymmetricGroup(14), [1..7], OnSets);; time;
1341

After changes:

gap> Orbit(SymmetricGroup(14), [1..7], OnSets);; time;
7

[ChrisJefferson]

This seems reasonable, and I agree the old hash function was very bad.

A couple of small suggested changes (just because it took me a while to review this, so it might be useful to other people!) These are very picky comments, but I read this fairly carefully because we are hashing bags.

Usually HashKeyWholeBag is unsafe, but in this case 'boundedtuples' will only ever be lists of integers. Maybe just add a comment saying that, for future readers worried about the HashKeyWholeBag?

Also, I'd make the last argument '1'. I don't think the 'Length(x)' is actually improving the hash value in practice, and I did assume it "meant something", rather than was just a seed.

[reiniscirpons]

Thanks for the comments! I've added a comment explaining that we expect the input to be a list of positive ints as well as an assertion to explicitly check this in the code. To avoid issues with different representations (like if the input were a range), we now also convert the input to a Plist if it is not already. Also agree with the comment about the seed value.

[ChrisJefferson]

Thanks, I'm happy with this now, but I'll let other people pop in and have an opinion.

[fingolfin]

Thanks. I've adjusted the title so it can be used as an entry in the release notes. As such I focused it more on the user facing effect.