Improvements

[Marcono1234]

Some minor improvements:

• Use Locale.ROOT for case conversion; toLowerCase() and toUpperCase() without explicit Locale use the system dependent default Locale
• Call trim() after having loaded JSON language models. This sounds quite simple but can safe multiple megabytes of memory.
• Directly store String in ngram map instead of wrapping it in Ngram.
• Don't use Regex to determine whether character is Japanese, instead use the respective UnicodeScript constants. This avoids calling toString() for every char and Matcher (during matching), and is less error-prone than Regex patterns which are not supported on some platforms (e.g. Android).
• Reduce object creations:
  • When splitting into words directly store the results in a list instead of first creating a StringBuilder and then splitting again.
  • Avoid calling String.slice(IntRange), this causes unnecessary memory allocations for the IntRange, see also Avoidable creation of range detekt/detekt#3823. Instead just call substring(Int, Int).

(Note that these are not the major performance improvements I mentioned recently, I will write an issue for that soon they are described in #101.)

[Marcono1234]

It looks also like the accuracy reports changed slightly but I am not sure if that is related to these changes.

[pemistahl]

Wow, thanks a lot for these extensive optimizations. I will need some time to review them but I will keep you posted soon.