Create better Tesseract trained data

The *eng.traineddata* isn't optimal for identifying Magic titles.
* The Æ for example isn't correctly identified.
* The 'F' in *Spires of Orazca* (transformed from *Thaumatic Compass*) is identified as 'r'.
* 'I' and 'J' in the new font is mistaken for 'l'.

We need to
- [ ] Create a better traineddata-file.
- [ ] Name it *mtg.traineddata* and replace *eng.traineddata* with it.

https://sv.wikipedia.org/wiki/%C3%86

Start by reviewing the existing eng.traineddata usage and the OCR workflow in this repository. Determine how Magic card title samples are collected and trained, then verify that a new mtg.traineddata replaces eng.traineddata and correctly recognizes Æ, the transformed F, and the new-font I and J.

Written by the indexing model from the issue text.