Porting datasets to new TorchData API

[fabrizio-ottati]

See #201.

List of datasets to be ported:

• NMNIST.
• STMNIST.
• NCARS.
• ASLDVS.
• CIFAR10-DVS.
• DVGGesture.
• NCALTECH101.
• POKERDVS.
• SMNIST.
• DVSLip.
• SHD.
• SSC.
• DAVISDATA.
• DSEC.
• MVSEC.
• TUMVIE.
• VPR.

[biphasic]

In the end I think we don't need to convert most of the datasets. For most classification datasets, one can just use an IterableWrapper like so

import tonic
from torchdata.datapipes.iter import IterableWrapper

dp = IterableWrapper(tonic.datasets.NMNIST('data'))
dp = dp.filter(lambda data: data[1]==0) # only get samples with label 0

I think where the new API might make the most sense is for DSEC and TUMVIE, but even for those, DataLoader2 is not super mature yet to justify a migration and if the new API is much more suitable for those, also IterableWrappers can be used.

[biphasic]

closing this for now because torchdata development is on hold pytorch/data#1196