luci-app-pbr-1.2.3: refresh stale service status after Save & Apply

[egc112]

Saving settings fires pbr's procd config.change trigger, which reloads the service asynchronously. LuCI reloads the page as soon as the apply completes, so getInitStatus() lands mid-reload and reports the service as stopped -- and nothing ever re-checked it, leaving a stale "Stopped" in the status box until the user refreshed the page by hand. The service control buttons already handled this via pollServiceStatus() in pbr/status.js; the plain settings-save path went through LuCI's generic form flow and had no equivalent.

Re-check the status from render(), but only while it looks like that transient state (enabled but not running), stopping as soon as it settles and giving up after ~90s. Intervals ramp 1.5s -> 3s -> 5s, so the common case flips within a couple of seconds without competing for CPU with the reload being waited on.

Gating on enabled && !running matters for cost: getInitStatus() is expensive on the router, since the rpcd handler builds a fresh pbr instance per request and so re-runs full platform detection plus an nft list table dump on every call. A normally-running service therefore costs no extra RPC calls at all.

Two deliberate choices worth recording:

• setTimeout rather than LuCI's poll module, matching pollServiceStatus() in pbr/status.js. A registered poll drives LuCI's global auto-refresh indicator, which then sits at "Paused" once unregistered, reading as though the page had stalled.
• The status box is re-rendered only once the state settles, not on every tick, so the service control buttons inside it aren't torn out from under the user while they are looking at a stopped service.

Known limitation: this covers stale "Stopped" only. If the page loads while the service is running and it later fails to come back, the box stays stale until a manual refresh -- the trade for zero polling in the healthy case. A service the user stopped deliberately also re-checks for ~90s before giving up, which is bounded and self-terminating.

[egc112]

Co-Pilot comment:

LGTM.

This reliably fixes the stale "Stopped" status after Save & Apply by re-checking status only when enabled && !running, with a bounded backoff and sensible ramping. Gating the checks to avoid extra RPCs and using setTimeout to avoid poll UI churn are both reasonable choices. No blockers from me.