Feedback UI improvements

[timokau]

Various improvements to the labeling UI. Main changes are event logging (mostly for keeping track of stats during labeling) and multi-rater support.

Overview:

• We now log rating events which enables stats display
• We support multiple raters with distinct rater ids
• Auto-skip already-annotated comparisons; resume sessions at first unannotated item on page reload. Save on Next, even if no trait was marked relevant
• Small UX changes:
  • Interactive index input for jumping to specific comparisons
  • Scroll-to-top when switching between comparisons
  • Configurable host/port for distributed annotation setups

[timokau]

CI failure was unrelated, I have pinned Gradio for now to fix it.