Use concrete field type for type stability

[yha]

Makes everything inferrable and faster.

[yha]

Tests seem to fail here, although they pass locally on my machine. I think the difference is in OffsetArrays version. I'll look into it.

[yha]

The problem is that JuliaArrays/OffsetArrays.jl#90 added a custom axis type to the OffsetAxis union, so the ambiguities resolved in #1 crop up again. This breaks existing functionality from #1 (nothing to do with this PR).
I think now the proper solution would involve fixing the way offset arrays are designed in Base, since the current design seems to basically guarantee that different packages defining similar methods will trigger ambiguities when used together.
For now I suppose we can either add a dependency on OffsetArrays and disambiguate on OffsetArrays.OffsetAxis, or add a version constraint for OffsetArrays (but can a version constraint be used without a dependency?).

[Vexatos]

I really wish there was a way to pull IdOffsetRange without having to add a hard dependency on OffsetArrays, at least for now. That is just really ugly and wasteful.

[Vexatos]

This PR looks good unrelated to the OffsetArrays issue, which I actually made an issue for now. Thanks for your contribution, and for looking into this.