Fix TPR byte-order handling for cross-architecture compatibility

Summary:
TPR  feature reads server_id from TCP options using raw memory cast, which works on x86_64 (little-endian) but would fail on ARM or other big-endian architectures. The wire format is little-endian but the parsing code didn't account for host byte order differences.

- On little-endian systems (x86_64): No change.
- On big-endian systems (ARM): Automatic byte-swap via __builtin_bswap32()

This diff should be No-op

Reviewed By: frankfeir

Differential Revision: D86576924

fbshipit-source-id: 472440efed0e760b1bd96fcb206521cbcce68307

Author: tagrawal03

katran/lib/bpf/pckt_parsing.h

@@ -201,7 +201,13 @@ __attribute__((__always_inline__)) int parse_hdr_opt_raw(
       return -1;
     }
 
+    // TPR wire format is little-endian. Read the value and convert to host
+    // byte order if needed for cross-architecture compatibility.
     state->server_id = *(__u32*)&tcp_opt[2];
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    // TODO (T244373617): Add stats for this case
+    state->server_id = __builtin_bswap32(state->server_id);
+#endif
     return 1;
   }