Use spawn for every mindsdb task that require torch

[George3d6]

Torch only likes running on spawned processes.

Ideally most processes should be forked instead but we can spawn mindsdb just fine, make sure that's done for every train/test/predict call and the results are properly returned.