ENH: Better API for collections instanciation

[OutSquareCapital]

It's common to initialize mutable collections (dict, list) without elements, with the intention to mutate it later.

Rust approach this with constructors like BinaryHeap::new, Vec::new, or even iter::empty.

Python approach this with None checks in list, dict, set, etc... at runtime (or empty tuples as argument, need to check).

The former allow maximal performance and clarity of intent, while the second is the most convenient to type (and since dict and list can be written as literals, CPython can optimize this under the hood, bypassing the potential performance trade-off).

EDIT: see advancements on this issue on the comment below

• We should check if there's a true speed difference between tuple default arg, None check, and MyCollection([]) calls.

[OutSquareCapital]

So the actual flexible constructors is already implemented for all concrete public classes were it made sense (except Heap's, Mappings and sorted collections).

For the specialized constructors, the base interface could look like that:

class FlexibleInit[T](PyoIterable[T], Protocol):
    @classmethod
    def of(cls, *elements: T) -> Self: ...
    @classmethod
    def from_iter(cls, iterable: Iterable[T]) -> Self: ...

class FlexibleWrapper[T, W](FlexibleInit[T], Protocol):
    @classmethod
    def wrap(cls, iterable: W) -> Self: ...

class FlexibleCollection[T](FlexibleInit[T], Collection[T], Protocol):
    @classmethod
    def new(cls) -> Self: ...

Notes

Iter does not has FlexibleInit, because new on a collection means: create a new container

While an empty iterator would just yield nothing infinitum, without creating any data structure to hold anything.

So new clearly means; no-op creation of an empty data structure, not: "give me x object with a specific behavior"

Remains to be seen if we, however, allow once on collections.