Difftorch

Support non-contiguous tensors

Improved `generic_jacobianv` and `generic_jacobianTv` to support functions with non-vector tensor inputs and non-vector tensor outputs

* Added operations that return the function's value in addition to the derivative. These are prefixed with `f`.
- `fgrad`
- `fgradv`
- `fjacobianTv`
- `fjacobianv`
- `fjacobian`
- `fhessian`
- `fhessianv`
- `flaplacian`
- `fdiv`
- `fcurl`
- `fdiff`
- `generic_fjacobianv`
- `generic_fjacobianTv`

* Added shorthands for the following operations:
- `g`: `grad`
- `gvp`: `gradv`
- `vjp` : `jacobianTv`
- `jvp`: `jacobianv`
- `j`: `jacobian`
- `h`: `hessian`
- `hvp`: `hessianv`
- `fg`: `fgrad`
- `fgvp`: `fgradv`
- `fvjp`: `fjacobianTv`
- `fjvp`: `fjacobianv`
- `fj`: `fjacobian`
- `fh`: `fhessian`
- `fhvp`: `fhessianv`

Added `generic_jacobianv`, `generic_jacobianTv` operations

Initial release