Feature request: show token throughput (tok/s) in the TUI

Is your feature request related to a problem?

The TUI doesn't surface generation speed anywhere, so there's no way to gauge inference throughput per response or compare models/providers without benchmarking externally.

Describe the solution you'd like

Show average tok/s (tokens per second) in the status line / session footer while a response is streaming, plus a final per-response average once complete (e.g. next to the existing token counts). A session-wide average (via a /stats-style command or footer) would also be useful.

Describe alternatives you've considered

Timing requests manually or benchmarking against provider APIs directly — neither reflects real usage inside opencode.