OptimizableBatch being removed in V2

[CompRhys]

I was wondering if the team would be willing to shared the thought process behind the removal of the OptimizableBatch in V2? Fairchem models seem to typically trade-off some speed/memory-footprint for higher accuracy and so I wondered if it's the case that there just weren't that many cases where the batching was beneficial in your workflows given current generation cards?

[mshuaibii]

Hi - It's unlikely OptimizableBatch is something we'll migrate over, especially given the recent torch-sim work. Naive benchmarking of these tests did not show notable improvements, and we did not spend much time optimizing this given our workflows do not leverage this.

We'd much more rather see V2 support in torch-sim than another implementation on our end. But open to feedback :)