Test SubTensorBridge thoroughly and fix any issues found

[bdezonia]

I am not sure that the SubTensorBridage class is correct or understandable. Maybe I need two arrays instead of three as inputs. The second array is the full coord specification of the origin (a combo of the last two long[] arrays of the inputs of the current implementation). Then also write some 3d tests with 2d slices in various directions (and forw/back orders if desired). Then find bugs and fix issues.