Redesign around IdOffsetRange

[timholy]

This is breaking and might be a good excuse to go to 1.0. Fixes the observation in #65 (comment) even for other array types that offset their axes (see this test).

The motivation is discussed in more detail in https://discourse.julialang.org/t/why-is-there-a-performance-hit-on-broadcasting-with-offsetarrays/32194. There's something pretty satisfying, though, about offsetting the axes by a "lazy" mechanism

OffsetArrays.jl/src/OffsetArrays.jl

Lines 18 to 21 in f383ae0

	struct IdOffsetRange{T<:Integer,I<:AbstractUnitRange{T}} <: AbstractUnitRange{T}
	parent::I
	offset::T
	end

OffsetArrays.jl/src/OffsetArrays.jl

Lines 41 to 42 in f383ae0

	@inline Base.first(r::IdOffsetRange) = first(r.parent) + r.offset
	@inline Base.last(r::IdOffsetRange) = last(r.parent) + r.offset

because it makes it more possible to preserve any custom behavior of the parent array's axes.

CC @mbauman

[timholy]

Test failure on julia-1.0 might be due to the absence of JuliaLang/julia#28941? We could either add a bound on the Julia version or figure out a workaround.