Rewrite select! to improve performance

[Ellipse0934]

As discussed in Issue #1861 . This will significantly improve performance for the select! function(over 1000x for 100,000 columns).
Passes the tests locally and I couldn't find any bugs upon testing.

[bkamins]

Looks good. Now when I look at it it can be even made shorter by using copy! and using the pattern discussed by @nalimilan in #1974.
So it would look like:

    copy!(_columns(df), _columns(df)[inds])
    x = index(df)
    copy!(_names(x), _names(df)[inds])
    for (i, n) in enumerate(x.names)
        x.lookup[n] = i
    end

[Ellipse0934]

Need to import Future.copy! for backwards compatibility with ver 1.0. Will do it tomorrow.

[bkamins]

Yes - exactly like in #1974 (so whichever gets merged first the other will need to rebase later).

[Ellipse0934]

Can you tell me why something like VERSION < v"1.1" && import Future.copy! is a bad idea ?
Future won't be unnecessarily imported. Behavior should be the same. Let me test it out now.
EDIT: Seems like the tests have passed.

[bkamins]

I think it is OK.
@ararslan - just please confirm if this conditional import is OK

[bkamins]

Ah - and I think we need to add Future to [deps] in Project.toml like in #1974 (again @ararslan - can you please have look and confirm?).

[Ellipse0934]

I think we are ready to merge.
Ping: @bkamins

[bkamins]

I am OK to merge this. Let us wait for @nalimilan or @ararslan to confirm so that we have a second pair of eyes :), as this is a fundamental function in the DataFrames.jl ecosystem 😄.

[bkamins]

@Ellipse0934 and @nalimilan - thank you for contributing to this. The change is really of order of magnitude improvement as @Ellipse0934 noted!