Add rules for training tools

[stweil]

For best accuracy of text recognition (the primary goal of OCR-D), additional training ("fine tuning") is needed. Therefore training tools should be part of an advanced OCR-D installation.

Building training tools requires additional dependencies and targets.

• Tesseract: dependencies for training, make training-install
• Calamari?
• Kraken? It is currently not in the default set of tools which are built, but might be useful for training, for example to create line images from text.
• Others?

Should we add rules? How to name the Makefile targets? Perhaps existing target names could be used with an added -training, like make all-training.