linear and index array accessors about gtensor

[germasch]

Fundamentally, I like a[{0, 7}]. I don't think it necessarily conflicts with a[100] since those can be disambiguated by the arg type, though one may have to be careful that there's no ambiguity with implicit conversions.

For now, I'd only support the sarray index to be of length == expr_dimension(E) that it's being passed to, so all it really does is unpack and pass things through to the multi-arg operator().

I'm not sure that linear indexing via something like a[100] is desirable to have (in the public interface). I'd say if one wants to have a flattened view, that'd better be supported by something like ravel or flatten. In terms of a cheap way to iterate over all values, one should probably provide iterators instead of some (seemingly) linear access.

On the data layout, I think it's another issue that will probably require some thought. Fundamentally, gtensor and gtensor_view use an array of strides, so the underlying data can be row-major or column-major or whatever, and access will work. I guess you can make a row-major gtensor right now using swapaxis. But I also think it's true that in some select places (other than the initial initialization of the strides) there is an assumption of col-major, ie., some expression stuff may break, and that should be handled properly, though I also think I remember some non-trivial obstacles to that...

[bd4]

Closed by #62 and #56. Implements the simplest form, in particular operator[] only supports shape types, not literal arrays. Linear access is handled via the flatten view helper.