Enable Crowdin workflow for firmware translations

[obrusvit]

In order to use Crowdin effectively, we need to modify the way we work with language JSONs.

• en.json (a source JSON) currently has the following KV pairs: str -> str. Change it to str -> str | dict where dict allows to use same keys but different strings for different UIs. These changes must be propagated into the generation of translated_strings.rs, python mocks and language blobs. This helps with:
  • setting strings of different lengths to different UIs
  • achieving de-duplication of various versions of the same string.
  • Something like this:

"title__signtx": "Sign transaction"

"title__signtx": {"Bolt": "Sign transaction", "Samson": "Sign tx", "Quicksilver": "Sign transaction"}

• a script that takes the source json of each language and splits it per UI layout, i.e.
  • en.json -> en_Bolt.json, en_Samson.json, en_Quicksilver.json
  • de.json -> de_Bolt.json, de_Samson.json, de_Quicksilver.json
  • etc.
• these JSONs are submitted to Crowdin firmware project
  • it's possible to set maximal lengths of strings for individual models
• after translations on Crowdin are finished, the resulting translated strings are merged together into the source JSONs
  • Optionally use Crowdin API to upload/download JSOns
  • Optionally perform sanity checks for consistency (e.g. set of keys is matching for all langs for a UI)