GPR minimiser has very high memory usage

**Description of the error**
When using a GPR minimiser, memory usage is significantly higher than the other minimisers for the same number of parameters and steps. For example, with 4 parameters and 5 steps MDMC is killed with an out of memory error (on a machine with 16GB of RAM), whereas GPO and MMC can handle 100+ parameters and 100 steps.

The memory usage comes down to line 302 in `MDMC/refinement/minimizers/GPR.py`:
`point_array = list(itertools.product(*predictive_coordinates))`
Here a product of predictive coordinates is constructed and cast to a list. Each array of coordinates contains 100 elements by default, so `point_array` quickly becomes very large.

**Suggested fix**
If possible the cast to `list` should be removed and `point_array` kept as an iterator. `point_array` is later passed to the GPR's predict method, so this may not be possible.
Alternatively, the predictions could be batched.