Discussion about the usage of torch.float64

[GiulioRomualdi]

I've noticed that in some parts, the fact that a torch type is float64 is forced. I think this can be problematic when we use Adam with other libraries.

A possibility is to remove the following lines

adam/src/adam/pytorch/torch_like.py

Lines 19 to 22 in a6a60bd

	def __post_init__(self):
	"""Converts array to double precision"""
	if self.array.dtype != torch.float64:
	self.array = self.array.double()

and

adam/src/adam/pytorch/computations.py

Lines 23 to 25 in a6a60bd

	gravity: np.array = torch.tensor(
	[0, 0, -9.80665, 0, 0, 0], dtype=torch.float64
	),

so the user can force the float64 with torch.set_default_dtype(torch.float64) as is done in the tests

[GiulioRomualdi]

cc @Giulero