r_to_py convertion for torch tensors

[dfalbel]

This allows converting between torch tensors and numpy arrays with no copy:

library(torch)

x <- torch_tensor(c(1,2,3,4), dtype = torch_int())

arr_from_ptr <- reticulate::py_run_string(
"
import numpy as np
def arr_from_ptr(pointer, typestr, shape, copy=False,
                 read_only_flag=False):
    buff = {'data': (int(pointer), read_only_flag),
            'typestr': typestr,
            'shape': shape}

    class numpy_holder():
        pass

    holder = numpy_holder()
    holder.__array_interface__ = buff
    return np.array(holder, copy=copy)

", convert = FALSE  
)

n <- arr_from_ptr$arr_from_ptr(as.numeric(x$storage()$data_ptr()), "i", 
                               shape = reticulate::tuple(4L))

print(n)
x$mul_(2L)
print(n)