Transpose confusion matrix to fix precision and recall

[CameronBodine]

Following up on discussion in #29, instead of changing the axis each metric is calculated on, simply transpose the confusion matrix to fix precision and recall calcs. This does change the value reported by Frequency_Weighted_Intersection_over_Union() also, but no other metrics.

I think FWIoU on the transposed confusion matrix is correct though as the frequency is calculated column-wise where each column is now the label, not the prediction.

[dbuscombe-usgs]

This looks good. Now the CM and all the metrics derived from it should be correct. Thank you!