Harden health probe against false restarts

Probe mpv-version (always present) instead of time-pos, which live
streams with caching off can report unavailable while playing fine —
that produced false unhealthy verdicts. Also require 3 consecutive
misses (~60s) before force-restarting a hung stream.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

internal/player/player.go

@@ -212,14 +212,17 @@ func (p *Player) Command(args ...any) (map[string]any, error) {
 	return nil, fmt.Errorf("no reply from mpv ipc")
 }
 
-// Healthy probes mpv over IPC and reports whether it is actively playing (has
-// a finite time position advancing). A false result signals the daemon to
-// consider restarting the slot even if the process is still alive (frozen).
+// Healthy probes mpv over IPC and reports whether it is responsive. It queries
+// a property that always exists (mpv-version) rather than time-pos, because a
+// live stream with caching disabled can legitimately report time-pos as
+// unavailable while playing fine — using it here caused false "unhealthy"
+// verdicts and needless restarts. A false result now means mpv's IPC didn't
+// answer at all, i.e. the process is genuinely hung.
 func (p *Player) Healthy() bool {
 	if !p.Running() {
 		return false
 	}
-	reply, err := p.Command("get_property", "time-pos")
+	reply, err := p.Command("get_property", "mpv-version")
 	if err != nil {
 		return false
 	}