remove rehash! from Dict constructor

[bkamins]

This PR follows https://discourse.julialang.org/t/is-there-a-bug-in-dict/6672/4.
It implements:

• Dict{K,V}(d::Dict{K,V}) constructor performs a bare copy of d (without rehash!);
• new test that currently fails and should pass after the PR;
• rename ht_keyindex2 to ht_keyindex2! (this is not strictly necessary for this PR so can be omitted, but it cleans up the related code, as ht_keyindex2 in some cases may call rehash! thus it may mutate the passed dictionary)

I recommend to remove rehash! from the constructor. Another approach would be to rehash a newly created Dict:

function Dict{K,V}(d::Dict{K,V}) where V where K
    d2 = new(copy(d.slots), copy(d.keys), copy(d.vals), d.ndel, d.count, d.age,
            d.idxfloor, d.maxprobe)
    d2.ndel > 0 && rehash!(d2)
    d2
end

but I do not see if there would be any huge benefit of this - rehash! will be called anyway if needed by setindex! or get! and it adds performance penalty during construction.

[stevengj]

Should we do dnew = new(...) and then rehash!(dnew) if d.ndel > 0?

[bkamins]

This is exactly what I have considered (as given in the comment to the PR). I have decided not to add it as it introduces a performance penalty for Dict creation. Probably @JeffBezanson can provide the guidance here.
Another approach would be to test d.ndel > "some value greater than 0" (i.e. if we have a lot of deleted entries we do rehash!, and if there are only a few we leave it without rehashing).

[bkamins]

As a second thought - my recommendation would be to remove rehash! from the constructor. Its main benefit is that later we do not have to rehash! both dictionaries twice. But this is something that we cannot avoid anyway.
If we believe that being able to a clean dictionary is an important feature I would rather recommend to:

• rename rehash! to _rehash!;
• expose a new rehash! as an exported function and define it for Dict and Set. This new rehash! should perform an additional check if the size is not decreased (as _rehash! is designed for growing only).

[stevengj]

Note that the rehash! was added by @JeffBezanson in 41f3cf8 ... I tend to think we should rehash! the newly created dictionary if needed, but I agree that we shouldn't modify the old dict.

[JeffBezanson]

Yes, this is ok as a (quasi-)bug fix. As a later enhancement, if the fraction of deleted keys is high we should probably copy the pairs by for (k,v) in d; newd[k]=v; end, rather than copying the whole structure followed by rehashing it.

[bkamins]

Good idea - I will do some benchmarks to find a reasonable break even and propose another PR.

[bkamins]

CI failure seems unrelated. Should this be merged?