tui: plan usage footer is not synced across TUI instances sharing one service #59062

Summary

With several OpenCode TUI tabs open behind a single shared background service,
each footer shows a different Go plan usage/allowance value, because the footer
only renders the opencode.allowance from the most recent inference response
that session completed. There is no polling of the shared service and no API
endpoint exposing plan allowance, so tabs show stale or divergent balances for
the same plan.

Environment

• opencode version: 2.1.4 (latest)
• OS: Linux 7.0.0-38-generic (linux x64) — Arch
• Terminal: alacritty (TERM=alacritty, COLORTERM=truecolor)
• Shell: /bin/bash
• Install/channel: latest
• Active plugins: @xberg-io/opencode-html-to-markdown

Reproduction

1. Connect a Go plan via /connect and use a Go model.
2. Open two or more terminal tabs running opencode in the same environment
   (all attaching to the same background service; verify opencode service status).
3. Send requests in one tab only.
4. Compare the plan-usage readout at the bottom of each tab.

Expected Behavior

All TUI instances connected to the same shared service should display the same,
current plan usage for the Go workspace — ideally refreshed periodically
(polling the shared service) rather than only updating on a session's own
inference response.

Actual Behavior

The footer in each tab shows only that session's last echoed allowance value.
Tabs not used recently (or brand-new sessions) show stale/initial values until
their next request. Nothing syncs while idle.

Additional Context

• Inspected the running service's /openapi.json: there is no endpoint exposing
  plan allowance. The closest is /api/experimental/session/stats, which returns
  aggregate token/cost stats per project — not plan state.
• The PlanTabs precedent: the Console's plan-tabs UI already uses a shared
  syncKey so all plan displays on a page stay synchronized; the same idea
  applies to TUI footers via the shared service.
• Workspace limits are enforced server-side per plan, so enforcement is correct;
  this is a display/observability gap only.
• Workaround: check usage in the OpenCode Console, or watch the local
  /api/experimental/session/stats endpoint for aggregate cost.