… but still clean up resources immediately (#3879)

## Description

closes #3867

relies on #3924

Alot of finagaling needed to happen to ensure that when the Endpoint is
dropped, it's underlying resources are cleaned up ASAP.

First, added a custom `Runtime` that wraps spawned tasks in a
`CancellationToken`. This allows for both graceful shutdown (which does
not use the cancel token, but instead waits for tasks to finish
naturally) and ungraceful abort (which uses the cancel token to stop all
running tasks)

Wrote a new `Handle::abort` method that is sync and does not wait
for graceful clean up after cancelling tokens, before aborting any runtime tasks.

In order to ensure that the `Endpoint` is only "aborted" on the
last clone of the endpoint, `StaticConfig` was moved into the
`Handle`, and it was renamed to `EndpointInner` that is now wrapped 
in an `Arc`. When the `Arc` is finally dropped, the `Drop` impl on 
`EndpointInner` is called, which calls `Handle::abort`, if the `Endpoint`
was not gracefully closed before dropping.

This PR also makes sure that we drop any references to address lookup
services on close.

## Breaking Changes

New behavioral change! Then `Endpoint` does not do ANY best-effort to
close connections gracefully before dropping. This makes it even more
important to call `Endpoint::close` in all cases.

This change (that prevents the underlying `quinn::Endpoint` to do it's
best to wait/send close frames before shutting down) has already
highlighted in tests how much more brittle it is to drop the endpoint
without calling `close` first.


## Change checklist
<!-- Remove any that are not relevant. -->
- [x] Self-review.
- [x] Documentation updates following the [style
guide](https://rust-lang.github.io/rfcs/1574-more-api-documentation-conventions.html#appendix-a-full-conventions-text),
if relevant.
- [x] Tests if relevant.

Co-authored-by: “ramfox” <“kasey@n0.computer”>
Co-authored-by: Franz Heinzmann <frando@unbiskant.org>