A way to iterate through directories in asset loading

## What problem does this solve or what need does it fill?

It's not so much of a feature request, but a report of a feature regression. As of bevy 0.11, it was possible to walk directories while loading an asset via `load_context.asset_io().read_directory(path)`. It was a handy way to dynamically collect other files and process them as labeled side-assets. It is now impossible to do that in bevy 0.12. And as I see it, such thing would probably be a better fit for the new asset preprocession than it was earlier, but as it stand now I see two issues:
1. I would still want to opt-out preprocessor for the time being (as I am to my shame still confused by it), and since it's a feature that is completely opt-outable, I would like a way to do things like I used to do it without preprocessing;
2. Even if I were to start using asset preprocessor, still, I think there is no way to replicate what I was doing the version before ? There is [`AssetReader`](https://docs.rs/bevy/latest/bevy/asset/io/trait.AssetReader.html#), but it is unavailable in the context of asset loading ?

## What solution would you like?

I am certainly a little bit lost with the new preprocessing, so for me it would be perfect to just get an access to `AssetReader` in the loading context, i.e:
```rust
fn load<'a>(
&'a self,
reader: &'a mut Reader,
_settings: &'a Self::Settings,
load_context: &'a mut LoadContext,
) -> BoxedFuture<'a, Result> {
Box::pin(async move {
let directories = load_context.asset_reader().read_directory("scripts").await?;
for item in directories {
// ...
}
// ...
})
}
```
This way I could upgrade parts of my codebase to bevy 0.12 without making much changes.

## What alternative(s) have you considered?

Preprocessing is a thing, and I am not totally sure if at this moment it is possible to replicate what I was doing using meta files and some clever processor.

## Additional context

I was using this feature in bevy 0.11 to build a project of teal scripts. It's a typed dialect of lua, so in the asset loader my main asset was a script project configuration, and labeled side-assets were scripts that I would transform from teal to lua. It was handy, because I didn't need to put every script into a list and save it in the configuration, it was dynamic and it was working quite well.

Read the referenced AssetReader trait, LoadContext, and asset-loader load entry point first, then compare them with the asset preprocessing path. Done means an asset loader can iterate a directory through its loading context and support the reported dynamic side-asset use case without requiring a manually maintained file list.

Written by the indexing model from the issue text.