Text layer recognition

[kirillt]

#176 and #178 really depend on resources having text layer attached when possible.

We can extract text relatively easily from:

• content of text-based resources
• metadata of some file types (JPG, MP3)

We can extract text with some effort from:

• images (e.g. screenshots of web pages, or photos with shop labels)
• audio (useful for voice-notes, which possible to make while driving a car)

Text layers should be stored in hidden folder, alongside with tags storage and index, and should be managed by arklib.