refactor(log): replace stdlib log with the application JSON logger

Every log.Printf/Fatalf becomes a structured slog call through
internal/log: request handlers use the request-scoped logger
(applog.FromContext, carrying request_id), startup failures exit via a
fatal() helper that still emits JSON, and the seedsample/dumpschema CLIs
bootstrap the same JSON logger. Stale client_test expectations aligned
with the refactored wrapper-call logging (message casing, result attr,
ERROR level).

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

backend/internal/garmin/sync.go

@@ -9,11 +9,11 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
-	"log"
 	"sync"
 	"time"
 
 	"geniusrun/backend/internal/classify"
+	applog "geniusrun/backend/internal/log"
 	"geniusrun/backend/internal/store"
 )
 
@@ -368,12 +368,12 @@ func (s *Sync) fillPendingWorkouts(ctx context.Context, limit int) error {
 				// after being linked to this activity) will never succeed on
 				// retry -- mark it so ActivitiesMissingWorkout stops
 				// surfacing it, instead of retrying forever.
-				log.Printf("sync: workout for activity %d not found on Garmin, marking as such (will not retry): %v", a.GarminActivityID, err)
+				applog.FromContext(ctx).Warn("workout not found on Garmin, marking as such (will not retry)", "garmin_activity_id", a.GarminActivityID, "error", err)
 				if serr := s.db.SetActivityWorkoutNotFound(ctx, s.userID, a.ID); serr != nil {
 					return fmt.Errorf("mark activity %d workout not found: %w", a.GarminActivityID, serr)
 				}
 			} else {
-				log.Printf("sync: fill workout for activity %d failed, will retry next sync: %v", a.GarminActivityID, err)
+				applog.FromContext(ctx).Warn("fill workout failed, will retry next sync", "garmin_activity_id", a.GarminActivityID, "error", err)
 			}
 		}
 		s.setProgress(PhaseWorkouts, i+1, len(pending))