Co-Authored-By: Nathan Sobo <nathan@zed.dev>

Blocking could happen while processing events, which would cause the
background scanner to never make any further progress.

This allows us to drop the context *after* we ran all futures to
completion and that's crucial otherwise we'll never drop entities
and/or flush effects.

This allows us to remove the dropped entities and flush effects as
necessary.

Co-Authored-By: Nathan Sobo <nathan@zed.dev>

This avoids a reference cycle that is causing some tests
to fail due to leaked handles at the moment. There may be
a better way to fix this though.

This probably isn't the *root* cause of why an executor is leaked,
but by cutting off this cyclic references, it may make it a bit easier
to track down leaks of an executor.

This bug was caught by running the executor until parked after tests.

Co-Authored-By: Nathan Sobo <nathan@zed.dev>

This way, at the end of a test, dropped entities will be removed, and will drop
the handles that they hold.

For now, just track models. Tests fail because we don't
yet clear the app contexts at the right time.

Co-Authored-By: Nathan Sobo <nathan@zed.dev>

Co-Authored-By: Antonio Scandurra <me@as-cii.com>

However, the randomized integration test is still failing:

```
ITERATIONS=100000 SEED=3027 OPERATIONS=200 cargo test --release test_random --package=zed-server -- --nocapture
```

Co-Authored-By: Nathan Sobo <nathan@zed.dev>

Blocking the sender could halt deserialization for no reason if nobody
is consuming the notifications.

...as we might be in the process of completing a request that could
open a buffer. This was causing a failure in the randomized integration
test.

This commits adds the beginnings of an application state facility as a non-static place to store the most recently-activated search for each project.

I also store workspace items by descending order of their entity id so that we always fetch the newest item of a given type when calling `Workspace::item_of_type`.

Just like the DOM, we now bubble events down the tree during a capture phase before bubbling them back up.

I thought I needed this but actually didn't, but I still kinda think it's a good change for the public interface of Editor.

We had some duplication when cloning on split, so this is to unify that before we add any more complexity to construction.

I'd love to fade the text out, but for now I just append an ellipsis.

Search is both a verb and a noun, which makes it more natural to use in situations where we need to name a thing rather than a process.

Also: Clear the selection when we focus the results editor because we continue to render the selection even when the editor isn't focused and it looks awkward. Another approach we could take is to not render selections for non-focused editors, either always or with an option. But considering that we select all anyways next time we return focus to the query editor, I think this is ok for now.

This replaces the `cmd-alt-shift-F` binding to open a new search. Instead, you can preserve the existing search results by entering a query and then hitting `cmd-enter` instead of `enter`. This opens a new project find view and restores the previous view to whatever query it was previously displaying. It's a bit strange, but I don't want to rely on splitting as the only way of creating multiple sets of search results.

Also: Wire up events so the modified status updates correctly.

Also: Introduce the ability to disable and enable the nav history directly. This allows us to explicitly push an entry when opening excerpts and then disable all pushes as we open individual buffers.

Also: Remove special handling for alt-shift-D binding in diagnostics view that opens excerpts. Rely on alt-enter in all multi-buffers instead.
Co-Authored-By: Max Brunsfeld <maxbrunsfeld@gmail.com>

Co-Authored-By: Max Brunsfeld <maxbrunsfeld@gmail.com>

Co-Authored-By: Max Brunsfeld <maxbrunsfeld@gmail.com>