Modifs to use custom models with omnipose version 1.0.6

[marieanselmet]

• Remove --verbose in cmd, as it causes Omnipose to crash when frames have no detections. Wasn't fixed in Omnipose version 1.0.6
• Add --nchan argument set to 1 to use custom models trained on single-channel images
• Add --nclasses arguments to be able to use custom Omnipose models with different nclasses values. By default, --nclasses is set to 2, as in the code of Omnipose for training branch. If the python process returns an error, we parse the python output error and guess the value for --nclasses in accordance (based on the tensor size of the biases for the last output layer of the model). We then build a new command and call a new python process with the updated command.
• Add a log message to the logger if the pretrained model path is incorrect