Added support for broadcasted piping .|>.

[racinmat]

Fixes #14 .
I added few tests, they are passing, but I'm not entirely sure that it doesn't break anything else.

[oxinabox]

what happens if we do:

_macroexpand( :(@pipe fn .|> _.(1:2) ) )

[racinmat]

Now it produces :(var"._".(1:2)) and @pipe fn .|> _.(1:2) crashes.
I'm not entirely sure, what is desired behaviour, how the resulting code should look like.
In plain Julia fn .|> x->x.(1:2) == fn |> x->x.(1:2), and since fn isa not iterable AFAIK, then fn ,|> produces same result as fn |>, so I'm not sure where I want to go with this.

[oxinabox]

we want to match the normal behavour.

Here is an interesting set of cases:

julia> [1,2,2] |> x-> atan.([10,20,30], x)
3-element Array{Float64,1}:
 1.4711276743037347
 1.4711276743037347
 1.5042281630190728

julia> [1,2,2] .|> x-> atan.([10,20,30], x)
3-element Array{Array{Float64,1},1}:
 [1.4711276743037347, 1.5208379310729538, 1.5374753309166493]
 [1.373400766945016, 1.4711276743037347, 1.5042281630190728]
 [1.373400766945016, 1.4711276743037347, 1.5042281630190728]

julia> [1,2,2] .|> x-> atan([10,20,30], x)
ERROR: MethodError: no method matching atan(::Array{Int64,1}, ::Int64)

[racinmat]

I see. If I understand it correctly, the embedding of input array is not able to cover cases 2 and 3, because the . notation is not expressive enough, so I should explicitly call broadcast function in the macro, right?
Because I can write test for

@test _macroexpand( :(@pipe [1,2,2] |> atan.([10,20,30], _) ) ) == :(atan.([10,20,30], [1,2,2]))

but I'm kinda lost in decising what should

@test _macroexpand( :(@pipe [1,2,2] .|> atan.([10,20,30], _) ) )

equal to.
What do you think? @oxinabox

[oxinabox]

Nice, can we replace every instance of the word elementwise with broadcast they are subtly different (depending how exactly one defined elementwise).

Other than that, and the comment i left re testing extra case.
this looked good to me.

Could you also bump the minor release number in the Project.toml so i can tag a release right after merging this?

[oxinabox]

can we also test

[true,false] .|> !
[1, 2] |> .+(_, x)
[1, 2] |>  _ .+ x)
[1, 2] .|> .+(_, x)
[1, 2] .|>  _ .+ x)

[oxinabox]

Maybe we can just detect if it is soing to attempt to pipe-broadcast a already broadcasted expression and say:
@pipe does not support broadcasted pipes into already broadcasted expressions. Rather than e.g xs .|> f.(a, _)you can use a normal lambda still:xs .|> b->f.(a, b)`

[oxinabox]

can we break this into two functions?
rewrite_apply and rewrite_apply_broadcasted

[oxinabox]

can we break this into two functions?
rewrite and rewrite_broadcasted

[oxinabox]

Looks all good to me, one last pair of suggestions then should be good to merge.

[racinmat]

So, I split them to separate functions, tests are passing, anything else, or is it ready to merge?

[oxinabox]

LGTM

[oxinabox]

thanks!