Redesign atomic/terminal elements

[jordens]

As discussed in #96 (comment) we could consider:

• Maybe we need to split our single trait into Miniconf and MiniconfAtomic for the following:
• Instead of a macro impl_single, blanket impl MiniconfAtomic for T where T: Serialize + Deserialize
• That extends MiniconfAtomic support to everything that is deserializable/serializable but makes them terminal/atomic elements.
• That then also covers array and give it atomic support
• Would also cover every struct that is Serialize + Deserialize
• Keep the current custom (inner) Miniconf (explicitly or derive) support for array, Option etc
• Add (inner) Miniconf support for Vec and others that are IndexMut + IntoIterator (?)
• Consider (inner) Miniconf for things that are container-like: tuples (macro impl up to 32 tuple size) or hashmaps.
• Add support for tagging specific struct members with miniconf(defer) (modulo bikesheding for a better name: defer, inner, non-atomic) to the derive(Miniconf) macro. This is along the lines how other crates like e.g. thiserror achieve fine grained control over derive macros (error(transparent) etc).
• That would allow deferring to the (inner) Miniconf impl within the member and make the member non-atomic.
• Otherwise and by default a struct member is exposed through its MiniconfAtomic.
• Maybe allow exposing both the outer and inner Miniconf explicitly in (the rare) cases where this is not an ambiguity problem.
• (Re)Consider supporting derive(Miniconf) for tuples, enums.
• This gets rid of the ambiguity of whether array/String/Vec are Atomic or not, i.e. whether they use inner or outer miniconf.

Not completely sure whether this fully maps onto the trait architecture of rust though and whether there are no conflicting trait impls or incompatibilities.
Not sure whether the traits Miniconf and MiniconfAtomic can/should/need to be disjoint for this.

#[derive(Serialize, Deserialize, Miniconf)]
struct Settings {
    a: [0u8; 2],  // "a = [0, 1]"
    #[miniconf(defer)]
    b: [0u8; 8],  // "b/1 = 1"
}