Move training process into tesstrain.sh system

[ryanfb]

Tesseract has introduced a new training system based around tesstrain.sh.

Ideally, we want to be able to submit a pull request against the langdata repo, similar to that for Ancient Greek: tesseract-ocr/langdata#19

This may also require modifying/pull-req'ing the lat section of the language-specific.sh script.

Right now work on this is happening in two places:

• the langdata branch of this repo
• the tesstrain branch of the tesseract_latinocr_docker repo

The plan is also to consolidate the latinocr-lattraining repo so that we can have a single top-level make rule which makes the files which will go into the langdata repo and calls tesstrain.sh for the final build (similar to the updated grctraining process for Ancient Greek), and tesseract_latinocr_docker will merely provide the OS environment and Tesseract install.

[ryanfb]

Forked language-specific.sh now at https://github.com/ryanfb/tesseract/tree/latin-language-specific