Adapt TUI speech prefetch to live synthesis

[hrhrng]

Summary

• keep Manual mode isolated from automatic token pacing across live transitions
• coalesce sentence seeks and isolate stale speech eras so repeated arrow navigation resumes reliably
• learn live synthesis throughput and feed it into the bounded playback-runway controller
• expose the rolling synthesis factor in the opt-in voice trace

Verification

• cargo test (full TUI suite passed; the real-device callback test also passed in isolation after reproducing its known timing flake)
• cargo clippy --all-targets -- -D warnings
• cargo fmt --check
• local release build installed atomically to ~/.local/bin/readio