Add Batched vdot to clean up fire implementation

[CompRhys]

Adds a batched vdot to torch_sim.math

[CompRhys]

@t-reents I attempted to copy the changes you added but appear to have failed to do so as I do not match your results.

[t-reents]

Interesting, at first glance it indeed seems to be very similar. Out of curiosity, do you also get better results than the previous version, but different ones compared to mine, or do you not even see any improvement?

One thing that I noticed while having a quick look, you also included the transformation of the forces according to the current deformation:

https://github.com/Radical-AI/torch-sim/blob/cc94facb1c164c8def7823a464cc7711fa956d97/torch_sim/optimizers.py#L1478-L1480

However, you don't use forces in the subsequent steps but rather state.forces. For example:

https://github.com/Radical-AI/torch-sim/blob/cc94facb1c164c8def7823a464cc7711fa956d97/torch_sim/optimizers.py#L1507

and

https://github.com/Radical-AI/torch-sim/blob/cc94facb1c164c8def7823a464cc7711fa956d97/torch_sim/optimizers.py#L1527

[CompRhys]

I saw results very similar to what's currently on main

[CompRhys]

updating to use forces everywhere wasn't a fix sadly

[t-reents]

I will have a more detailed look tomorrow, already dedicated some time for it. I'll also double once again check if there are any additional changes on my side that I didn't properly, but in principle it shouldn't be the case.

[t-reents]

Already had a quick look. I can reproduce my results (on my branch, so what I shared should be complete). Moreover, when I just checkout your branch and run the script, I get the same results as on my branch, see attached screenshot. Based on this test, it seems that you reproduce my results, doesn't it?

[CompRhys]

closing in favor of #203!