Updated coincident to its latest v4

[WebReflection]

This MR brings in my latest (I promise) attempt at fixing/improving/testing coincident and its FFI capabilities.

The lengthy version of the reason I am updating this again is here but it can be contextually summarized as such:

• stability: it was nearly impossible to test/track possible issues behind the remote ffi scene, now it's a 100% test covered project with zero dependencies that can easily be augmented, improved or fixed, if needed
• features: I had a request around Float32Array being deadly slow with current PyScript dance and it was coincident to blame ... now all conversion are done properly and views or buffer are first-class citizens of this project (from anything worker related). On top of that, after discussing with @ntoll a little while ago I agreed there was room for improvements with dedicated APIs that helps reducing roundtrips whenever that's desired and/or possible
• maturity: coincident v3 (it lasted 2 months) was not really clean in features and purpose + it was leaving too much responsibility to the user with 4 variants of its internal encoder/decoder, helping literally nobody (included myself) to focus on one way to do things, a way that already works out of the box no matter what. Accordingly, I've squashed into a single, general purpose, always fast and simple encoder and decoder to forget forever about recursion, non-supported JS types and whatnot ... that is, the project behind has the goal to forget entirely about that choice and just work, which is currently validated by its 100% code coverage for all meant use cases and complex references that should be able to travel as fast as possible between "worlds"

/cc @ntoll here to align but I hope this new, and hopefully latest, version of that part of the stack, also now better documented and tested, is actually welcomed.