support permutedims and reshape

one common thing is to using broadcasting on named vectors, for which one is put into another dimension.

permutedims(a) as well as reshape(a, 1, :) can be used to bring a column vector a to row format. reshape(a, 1, 1, :) would create a 3rd-dim vector.

It would be great if these operations preserve the indexing.

using AxisArrays
AxisArray(1:3, ["A", "B", "C"])["A"] # works and returns 1
permutedims(AxisArray(1:3, ["A", "B", "C"]))["A"]

gives an error

ArgumentError: invalid index: "A" of type String

Stacktrace:
 [1] to_index(::String) at ./indices.jl:297
 [2] to_index(::Base.ReshapedArray{Int64,2,AxisArray{Int64,1,UnitRange{Int64},Tuple{Axis{:row,Array{String,1}}}},Tuple{}}, ::String) at ./indices.jl:274
 [3] to_indices at ./indices.jl:325 [inlined]
 [4] to_indices at ./indices.jl:322 [inlined]
 [5] getindex(::Base.ReshapedArray{Int64,2,AxisArray{Int64,1,UnitRange{Int64},Tuple{Axis{:row,Array{String,1}}}},Tuple{}}, ::String) at ./abstractarray.jl:980
 [6] top-level scope at In[182]:1

The ultimate goal would be that the following works and preserves indexes as would be intuitive:

a = AxisArray(1:3, ["A", "B", "C"])
tuple.(a, permutedims(a))
# currently returns a plain standard Array
3×3 Array{Tuple{Int64,Int64},2}:
 (1, 1)  (1, 2)  (1, 3)
 (2, 1)  (2, 2)  (2, 3)
 (3, 1)  (3, 2)  (3, 3)

and similar for using reshape instead of permutedims

Start by running the issue's AxisArray, permutedims, reshape, and broadcast examples to reproduce the indexing errors and plain-Array result. Trace the array transformation and indexing entry points involved, then verify that permutedims, reshape, and the demonstrated tuple broadcast preserve axis indexes without breaking existing behavior.