Tensor element getter and setter definition
related to #28 (closed) and #58 (closed) Tensor element getter and setter have an interface accepting coordinates or a flat index.
As mentioned in #58 (closed), flat index is ill-defined so far.
Besides with future Tensor that can be view/extract of larger Tensor, it would be technically possible to get/set values inside the bigger one, from the included one.
Shouldn't we restrict getter and setter to work only on the area exposed by the Tensor (not the possibly larger one that would actually be in memory)? If yes, the flat id used could be specified as the lexicographical index inside the exposed area (see #58 (closed)).
NB with current Tensor implementation the issue does not apply as there is no difference between exposed and stored tensor. So deciding on this issue will have no consequence on existing code, only on upcoming versions (see #42 (closed)).
NB my proposal is to restrict.