Deal with double overflows when ndsize_t elements need to be represented

For example, when accessing elements of a dimension, which is of type `ndsize_t`/`NDSize`, calculations can in theory overflow the precision and we thus "lose" information. Currently such situations are silently ignored. The question is how to deal with such cases.

Known occurrences:

- [ ] [`src/Dimensions.cpp`](https://github.com/G-Node/nix/blob/72d4e357766d78dafe52521786453793ee1b96a5/src/Dimensions.cpp#L235)