Talkinghead next3

[Technologicat]

• Improve talking animation, looks at least serviceable for now.
  • There's still a small delay at the client end, after the AI starts writing and before the animation starts, but I haven't yet figured out the cause.
• Decouple animation speed (w.r.t. walltime) from render FPS.
  • Now if rendering is slow, the animator effectively skips frames (like most video games do) instead of slowing down to a crawl.
  • Only the pose interpolator needed a nonlinear scaling applied to its step size; all other animation drivers (blinking, talking, breathing) were already defined in terms of wall time.
• Code cleanup.
  • Refactor model installer.
  • Division of responsibilities: set_emotion, set_emotion_from_classification.
  • Use consistent API endpoint naming scheme for the server functions implementing the endpoints.
  • Add some docstrings.
  • autopep8 server.py, and also manually make the code into more idiomatic Python (no more flake8 warnings).
  • This is isolated to commit 6633535, just in case if it's not desirable.

[Technologicat]

Also:

• Fix the saving of the eye_unimpressed morph.

It has always had two morph targets in THA3 (left and right), but the JSON has only had one "eye_unimpressed" key.

This is technically a breaking change, but prior to the version of talkinghead being developed [in this series of PRs], the emotion JSON file format was not part of any public API. There was just a bunch of ready-made templates that were used internally by the live plugin. However, now that we have a working manual poser [and especially since I'm going to make the emotion templates configurable per-character], we can assume that from the next release on, emotion JSON files will exist in the wild.

Therefore, I propose to fix this now (as in commit https://github.com/SillyTavern/SillyTavern-Extras/pull/214/commits/a794ad9dde5517e66fd64901dc503209568d718e), and only consider the emotion JSON file format part of the public API from the next release onward.

[Technologicat]

Added one more fix - forgot about the per-frame probability trigger in the blink animation. Oops. Now that's scaled properly, too.

@Cohee1207: Now this PR is final, any upcoming changes will go into a new branch. Comments?

[Cohee1207]

Nice, good work. Appreciate the math-heavy comment on the FPS algorithm.

[Technologicat]

Nice, good work. Appreciate the math-heavy comment on the FPS algorithm.

Just my usual approach when writing numerical solvers. This time the math was really easy, though, and we got a simple analytical solution :P

Thanks for the merge!