fix outdated choose_multiple_weighted docs

[TheIronBorn]

Seems we missed this when switching to select_nth and when fixing the complexity documentation

[dhardy]

👍

The odd thing is that sample_efraimidis_spirakis has a special case for length > (core::u32::MAX as usize), but IIUC that case would use at least 64 GiB of memory and be quite slow.

In any case, the implementation cannot be faster than O(length) time since the sum of weights is provided (unless the user was able to provide that directly).