clarify restrictions on axes for general interface

[tpapp]

I am opening the issue following this discussion about the interface for general indexing.

The manual on Interfaces and custom indices specifies that axes should return (tuples of) AbstractUnitRange{T}.

In principle this would allow any T for which oneunit is defined, eg Dates or a wrapper type for floats (I have an example in the discussion I link above). This makes it tricky when one wants to perform arithmetic on indexes, in cases where a simple loop is not sufficient.

OTOH, if the implicit assumption is that T <: Integer, as @mbauman suggested in the discussion, it would be great if the manual could clarify that.

[tpapp]

Another clarification that could be added to the docs of axes is that axes themselves could use generalized indexing.

This is currently the case with OffsetArrays.IdOffsetRange, eg the following code in general will not work:

function all_same(v::AbstractVector)
    v1 = first(v)
    for i in axes(v, 1)[2:end] # WRONG: use (begin+1):end
        v[i] == v1 || return false
    end
    true
end

but this may trap an unwary user.

@timholy, what do you think?