Major refactor and update for 1.0

[mbauman]

Use a new lazy iterator forumalation to avoid excessive allocations (a la Base.LogicalIndex). Fixes #2 and should fix #1 (although I've not run any benchmarks yet).

[yurivish]

This is great!

Is there any way to make custom index packages like this one work with arrays that have custom indexing?

I've been playing around with my own custom array type trying to figure out how to make it work, but then noticed that AxisArrays also doesn't support them:

julia> using AxisArrays, InvertedIndices

julia> B = AxisArray(reshape(1:15, 5, 3), .1:.1:0.5, [:a, :b, :c]);

julia> B[:, :b]
5-element AxisArray{Int64,1,Array{Int64,1},Tuple{Axis{:row,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}}}}:
  6
  7
  8
  9
 10

julia> B[:, Not(:b)]
ERROR: ArgumentError: index InvertedIndex{Symbol}(:b) not found
Stacktrace:
 [1] axisindexes at /Users/yurivish/.julia/packages/AxisArrays/G6pZY/src/indexing.jl:290 [inlined]
 [2] axisindexes at /Users/yurivish/.julia/packages/AxisArrays/G6pZY/src/indexing.jl:174 [inlined]
 [3] macro expansion at /Users/yurivish/.julia/packages/AxisArrays/G6pZY/src/indexing.jl:354 [inlined]
 [4] to_index(::AxisArray{Int64,2,Base.ReshapedArray{Int64,2,UnitRange{Int64},Tuple{}},Tuple{Axis{:row,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}},Axis{:col,Array{Symbol,1}}}}, ::Colon, ::InvertedIndex{Symbol}) at /Users/yurivish/.julia/packages/AxisArrays/G6pZY/src/indexing.jl:311
 [5] getindex(::AxisArray{Int64,2,Base.ReshapedArray{Int64,2,UnitRange{Int64},Tuple{}},Tuple{Axis{:row,StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}},Axis{:col,Array{Symbol,1}}}}, ::Function, ::InvertedIndex{Symbol}) at /Users/yurivish/.julia/packages/AxisArrays/G6pZY/src/indexing.jl:117
 [6] top-level scope at none:0

NamedArrays implements its own Not here.

[mbauman]

If AxisArrays properly overloaded (and called) Base.to_indices instead of defining its crufty old to_index function it would just work™.

This PR asks to_indices to provide the proper index type for the skipped indices, so this will play nicely with any package that overloads that API. Then the indexing definitions in other packages will need to call to_indices themselves if they accept ::Any index types. That's the new hotness and how I'd recommend doing things these days. I explicitly designed Base.to_indices with the explicit intention of someday using it in AxisArrays... just hasn't happened yet.

[yurivish]

Awesome. Thank you!