Make indexin first argument accept any iterable

[garrison]

I also fixed the docstring and added a test where a matrix is passed.

Also compactified the tests according to https://github.com/JuliaLang/julia/pull/8622/files#r18615289 :-)

[garrison]

Actually, with this change, the first argument to indexin can now be a scalar as well:

julia> indexin(4, [3,4,5])
0-dimensional Array{Int64,0}:
2

but notice that it returns it in the form of a zero-dimensional array. Should this call return an Int instead?

[StefanKarpinski]

Should this call return an Int instead?

Yes, 100%.

[garrison]

OK, I split this into two commits: the first updates the docstring and tests according to the old functionality, while the second adds new functionality.

I'm actually unsure the best way to implement the goal of having a scalar return a scalar instead of a 0D array. My current approach has one broken test (marked by @test_broken) that I would claim ought to work if this were implemented correctly. Also, the docstring is now inaccurate since the method no longer always returns an array.

[garrison]

I fixed the 0-dimensional array issue by changing from a comprehension to calling map instead. I also decided that IHMO, having the docstring refer to an array is appropriate since the most common use involves returning an array, not a scalar.

One remaining issue is that indexin(Dict(3=>4, 5=>6), [1,2,3,4]) currently throws a BoundsError because Dict cannot be indexed by "positions" 1 and 2. But this actually calls a deprecated method (map(f, d::T) where T <: Associative), so it will instead throw MethodError in the future, which is more appropriate. At the moment I'm trying to determine if there may be other types that would fail in less graceful ways.

[garrison]

I'm pretty happy with how this stands -- I don't think the Dict issue I mentioned above is a big deal. If desired, I could always check that eachindex(b) == 1:length(b) (and throw an ArgumentError if it does not).