Aurelia Molzer

Fyi, the bulk of this PR looks good. But, from the perspective of _secure defaults_, I don't yet feel comfortable with an interface doesnt provide *some* secure configuration of all...

The approach is decent with the trait factoring, but I can't quite get over the _modification_ to `new` and other constructors instead of a `with_` method and `new` being dependent...

It might be possible but the field `Matches::free` has type `Vec` so any additional path argument can't be collected into it without being valid UTF8. It would be necessary to...

This might be resolved now. `ArcRef` is `Clone` which allows one to duplicate the handle and then map the two instances independently into the sublisces.

> How to handle async? The strategy can be: don't. What the caller more precisely wants to do is to construct a `Future`. There are several reasons why this is...

> Do you see any problem with it? This is how you can write a specific Future type, and indeed this is possible, but it can not be a trait...

There is a conflict of interest here, which also becomes apparent in the custom wrapper for `MaybeUninit`. Either the crate offers `std`-like trait in a `no_std` environment, with good compatibility....

Just to clarify, PIL uses a C version indirectly https://github.com/python-pillow/Pillow/blob/ab9a25d623fdd7f8de3e724b538f5660eac589ae/src/libImaging/BoxBlur.c#L294 It's still somewhat embarrasingly slow.

Interesting idea. Maybe we should deref into the types of the [`rgb`](https://docs.rs/rgb/0.8.25/rgb/) crate instead of defining our own? It's possibly one of the most stable crates for the domain and...

> It would take more discussion but I do think a strong case could be made for adopting the standard rgb crate types outright (though perhaps wait for them to...