change call scheme for any and all for AbstractArray

[matthias314]

This PR makes three changes to the way any and all calls are transformed into 3-argument _any and _all calls for AbstractArray:

• any(a) is transformed to any(identity, a) instead of calling 2-argument _any, and analogously for all. I think this has the following advantage: If somebody wants to implement any for a custom AbstractArray type, then it is now enough to define the 2-argument any. Currently also the 1-argument version has to be defined explicitly.

• The restriction that the predicate f for the 2-argument versions of any and all be of type Function has been removed. None of the other methods for the 2-argument versions of any and all impose this, and I see no reason to have it here.

• I've deleted the explicit definitions of the 2-argument versions of _any and _all. They are covered by the following line
  

  julia/base/reducedim.jl

  Line 1007 in 39f2ad1

  $(_fname)(A, dims; kw...) = $(_fname)(identity, A, dims; kw...)

  
  in the for loop a few lines later. (I believe that it doesn't matter anyway. As far as I can tell, the 2-argument _any and _all were only used for the 1-argument any and all and nowhere else.)

[mbauman]

This change looks like a great idea!

I don't fully understand (yet) why any and all are defined differently than the above loop... if there is a reason to it at all. It seems like they could — and should — all be defined in exactly the same manner. And then it'd be nice to effect this change for all the reduction functions by re-jiggering the loop above in the same way you did here.

[matthias314]

Do you want to move forward with this? If yes, do you want changes for other reduction function to be part of the same PR or separate? (I would prefer the second option.)

[jishnub]

It's probably better to do these separately