• Bump piper floor to 1.8.0 for the Japanese and Thai phonemizers
• Add ja (OpenJTalk) and th (TLTK) optional dependencies, with matching
  script/setup --ja and --th flags
• Add ja to the Docker images. th is left out: TLTK pulls in gensim,
  scikit-learn, scipy and pandas for ~500 MB, so install the extra by hand
• Bump wyoming floor to 1.10.2
• Bump the sentence-stream floor to 1.2.1 to match piper-tts[zh]
• Stop installing torch in the default Docker image. Piper has not needed it
  since 1.6.1 trimmed the zh extra; it is now installed only for omnivoice