fix(reassembly): unfragmented datagram with a buffered BUFID is dropped and flushes IPv6 #2270

Describe the bug
In IP.reassembly, a datagram with FO 0 and MF clear whose BUFID is already buffered submits the old buffer and returns. The datagram itself is dropped. IPv6 inherits this, so a colliding atomic fragment also cuts short the datagram being reassembled.

Reproduction (Packet records, BUFID shared, strict=False; strict=True behaves the same)

IPv4: #1 FO=0 MF=1, #2 FO=0 MF=0      -> [((1,), PARTIAL)]          #2 alone -> [((2,), COMPLETE)]
IPv6: #1 FO=0 M=1, #2 atomic, #3 FO=16 M=0 -> [((3,), PARTIAL), ((1,), PARTIAL)]

Expected behavior
IPv4: RFC 791 §3.2 flushes the BUFID and then submits this datagram, so (1,) PARTIAL plus (2,) COMPLETE. IPv6: RFC 6946 §4 says an atomic fragment MUST be processed in isolation, and the queued fragments MUST NOT be discarded. That gives (2,) COMPLETE and (1, 3) COMPLETE. Suggested fix: after the flush, fall through and reassemble the IPv4 datagram on its own. On IPv6, complete the atomic fragment without touching self._buffer.

System information
Linux, CPython 3.14.9, pcapkit 1.5.0b14, checkout main 9d504ead5.