refactor: replace panic calls with structured error handling

Replace ~25 panic() calls in non-fatal code paths with proper
s.logger.Error + return patterns. Panics in handler code crashed
goroutines (caught by defer/recover but still disruptive) instead
of failing gracefully.

Key changes:
- SJISToUTF8 now returns (string, error); all 30+ callers updated
- Handler DB/IO panics replaced with log + return/ack fail
- Unhandled switch-case panics replaced with logger.Error
- Sign server Accept() panic replaced with log + continue
- Dead unreachable panic in guild_model.go removed
- deltacomp patch error logs and returns partial data

Panics intentionally kept: ByteFrame sentinel, unimplemented
packet stubs, os.Exit in main.go.

server/channelserver/handlers_cafe.go

@@ -96,7 +96,8 @@ func handleMsgMhfGetCafeDuration(s *Session, p mhfpacket.MHFPacket) {
 	var cafeTime uint32
 	err = s.server.db.QueryRow("SELECT cafe_time FROM characters WHERE id = $1", s.charID).Scan(&cafeTime)
 	if err != nil {
-		panic(err)
+		s.logger.Error("Failed to get cafe time", zap.Error(err))
+		return
 	}
 	if mhfcourse.CourseExists(30, s.courses) {
 		cafeTime = uint32(TimeAdjusted().Unix()) - uint32(s.sessionStart) + cafeTime