Transform propagation sometimes takes a very long time (300ms+)

## Bevy version

0.16.0

## Relevant system information

cargo 1.89.0-nightly (47c911e9e 2025-05-14)
Windows 11

## What you did

Sometimes we move a lot of objects around (floating origin) and need the transforms to be updated for later systems. If we need to move everything we call.

```rust
commands.run_system_cached(bevy::transform::systems::propagate_parent_transforms);
```

After some trial and error and reading the docs, we now call:

```rust
commands.run_system_cached(bevy::transform::systems::mark_dirty_trees);
commands.run_system_cached(bevy::transform::systems::propagate_parent_transforms);
commands.run_system_cached(bevy::transform::systems::sync_simple_transforms);
```

Sadly i was unable to find in which order these should be called.

In 0.15.0 we only scheduled `propagate_transforms`.

## What went wrong

After upgrading to bevy 0.16.0 transform propagation sometimes takes very long (even 300+ms). In 0.15.0 the transform propagation wasn't even something we noticed in tracy.

## Additional information

Screenshot of tracy displaying the time distribution, notice the stragglers at the right.

These seem to coincide with the long running transform propagation. Seems odd to have so many threads taking so long.

Example of a scene: