Fix #394 with the missing keys after sorting

[oxinabox]

This is my attempt at fixing #394.
Using idx = ht_keyindex2(d, key) or idx = ht_keyindex(d, key, true)
fixes the problem with the key vanishing,
but causes problems with out of bounds errors.

I tried a few other things.
Then I took a look at how rehash! worked,
and saw that it rebuilds the slots based only keys and values (if nothing has been deleted since last rehash!)
And so I figured: Just correct the keys and values and then rehash it.

I have also created a Julia0.6 branch with this change.
https://github.com/JuliaCollections/DataStructures.jl/tree/julia0.6
What I should have done is created a 0.6 branch without this change then PRed that.
So if this PR gets rejected cos it is wrong, that branch should be deleted.
Otherwise, that branch should be used to tag v0.8.4

@timholy this does not need to be ported to OrderedCollections.jl (I tried without looking by just stash applying the change set, and was informed that) OrderedCollections.jl is missing sorting functionality and that corresponding file.
I think that file should be moved over.
But that can be done with this change applied.

[codecov]

Codecov Report

Merging #395 into master will decrease coverage by 0.34%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #395      +/-   ##
==========================================
- Coverage   98.81%   98.47%   -0.35%     
==========================================
  Files          30       30              
  Lines        1777     1771       -6     
==========================================
- Hits         1756     1744      -12     
- Misses         21       27       +6

Impacted Files	Coverage Δ
src/dict_sorting.jl	90.9% <100%> (-1.4%)	⬇️
src/sorted_dict.jl	98.16% <0%> (-1.84%)	⬇️
src/ordered_dict.jl	92.48% <0%> (-1.84%)	⬇️
src/priorityqueue.jl	98.96% <0%> (-1.04%)	⬇️
src/deque.jl	98.23% <0%> (-0.07%)	⬇️
src/circ_deque.jl	100% <0%> (ø)	⬆️
src/sorted_multi_dict.jl	100% <0%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 437f0bf...ebffdd1. Read the comment docs.

[StephenVavasis]

I'll just remark that I proposed in #382 and #386 a complete rewrite of OrderedSet and OrderedDict in terms of Array and Dict (rather than in terms of low-level Julia primitives as in the current code). The purpose of the rewrite is to improve maintainability. The rewrite removes some undocumented functionality, namely, use of integer subscripts on OrderedSet, which appeared to be incompletely implemented anyway.

[oxinabox]

@StephenVavasis i do not thing this is mutually exclusive with that.
Either here, or in OrderedCollections.jl

In that this is (I think) simpler than that,
and so can be merged first while you are sorting out the re-basing (and maybe the rebalancing )you mentioned in the last comment.
And then overwritten with your change once that is ready to go.
(While keeping the new test)
If that is the way we are going to go with this.

However::
I recall that I didn't venue to say anything on that thread because I concluded that I had only a limitted understanding of how HashTables worked,
and even less how your SortedDicts etc worked.

As such I didn't venture and opinion there, and I still don't.
Honestly my understanding of if this fix even works is limited, and kind of trial and error.
It passes all the tests.

[timholy]

The rewrite removes some undocumented functionality, namely, use of integer subscripts on OrderedSet, which appeared to be incompletely implemented anyway.

I've already deprecated it in OrderedCollections, see JuliaCollections/OrderedCollections.jl@35ac721.