Add SparseLatticedTensor

[ValerianRey]

TODO:

• [ ] Add a special type of error for that, like UnsupportedOnSparseError. We could catch that + NotImplementedError, but not catch ValueError for instance.
• [ ] Prioritize using aten (low level) operations over torch (high level) operations to avoid going through all the dispatch mechanics several times.
• [ ] Add pointwise and 0-to-0 functions that take one Tensor and some other arguments that can be directly forwarded to the call on the physical to the default implementation of pointwise functions.
• [ ] Be careful about composition, need to make sure that every operation returns the right type of physical (and need tests).
• [ ] Investigate overflow of the hnf_decomposition algorithm, with a matrix of shape [5, 7], rank 3 and values in range [-50, 51[, there was overflow (See test prior to ba9bf21762d55483fbc06670698e78307e2fb725)