feat: classification reads max HR from the profile instead of static config

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

backend/internal/sync/service.go

@@ -35,8 +35,6 @@ type Config struct {
 	// MinConfidence is the classify.Classify threshold below which even a
 	// single matching kind is sent to manual review.
 	MinConfidence float64
-	// MaxHR is used only to derive avg_hr_pct_max for the rule engine.
-	MaxHR float64
 }
 
 func (c Config) withDefaults() Config {
@@ -296,8 +294,16 @@ func (s *Service) ClassifyActivity(ctx context.Context, activityID int64) error
 	if err != nil {
 		return fmt.Errorf("parse workout kind rules: %w", err)
 	}
+	profile, err := s.db.GetProfile(ctx)
+	if err != nil {
+		return fmt.Errorf("load profile: %w", err)
+	}
+	var maxHR float64
+	if profile.MaxHeartRate != nil {
+		maxHR = *profile.MaxHeartRate
+	}
 
-	ctxMetrics := buildMetricContext(activity, laps, s.cfg.MaxHR)
+	ctxMetrics := buildMetricContext(activity, laps, maxHR)
 	result := classify.Classify(ctxMetrics, rules, s.cfg.MinConfidence)
 
 	candidatesJSON, err := json.Marshal(result.Candidates)