Memory Profiling Support

## What problem does this solve or what need does it fill?
The `3d_scene` example currently uses 107.5 MB of resident memory on my machine, even with all of the optimizations enabled. For a simple rendered scene of a box and a plane, I'd expect that to be lower. To better address this, it's important to be able to profile and monitor the memory behavior of Bevy apps.

## What solution would you like?
Either better docs or tooling for the following:

- Live memory snapshotting: finding what is allocating, how big are the allocations, what feature is it tied to.
- Allocation behaviors: When are allocations being allocated, when are they being freed?
- Cache behaviors: L1/L2/L3 cache hit rates.

For best platform support, this ideally should also support doing this remotely, though this might be dependent on platform specific tooling.

## What alternative(s) have you considered?
Blindly start shrinking data structures and cutting dependencies until memory usage goes down.