Keep original Scene -> Node Mapping in GLTF Asset Loader

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

I want to build a GLTF viewer that shows both the scene as well as the tree structure of a given GLTF scene.

For that to work I need to know which nodes are the basis of a scene.
However currently all nodes just get added to the bevy Scene and it is impossible to tell which nodes created that scene.
https://github.com/bevyengine/bevy/blob/1d0d8a3397bd6fc2c14d42ffd0668d2443748912/crates/bevy_gltf/src/loader.rs#L283-L294

## What solution would you like?

I would like to keep a field `scenes_to_node: HashMap, Vec>>` in the Gltf file
that preserves this information.

## What alternative(s) have you considered?

I could load the file with the [gltf::GLTF](https://docs.rs/gltf/latest/gltf/struct.Gltf.html) struct myself, get the indices of the nodes and then look them up in the `nodes` field of [`bevy::gltf::Gltf`](https://docs.rs/bevy/latest/bevy/gltf/struct.Gltf.html).

However this would require the program to load the file twice (and likely one time blockingly).

## Additional context

I believe the implementation to be fairly trivial. If there is support/acceptance for this feature I'm willing to implement it myself, however I first wanted to reach out here.

Start in crates/bevy_gltf/src/loader.rs at the linked scene-loading code, then inspect the Gltf asset's existing scene and node handle fields. Preserve the original scene-to-node relationships in the proposed mapping, and verify that each loaded scene identifies the nodes that created it.

Written by the indexing model from the issue text.