fix(reassembly): TCP's gap cap (TCP.__window__) has no RFC basis and silently splits a stream #2546

From the 2026-11-01 conventions audit, re-verified on 73add05a3. reassembly-integration.rst, "The Rule" point 4: reassembly follows the RFC's own definition and "adds no timeout, cap or ordering rule the RFC does not call for".

• TCP.__window__ = 1 << 24 (16 MiB) in pcapkit/foundation/reassembly/tcp.py, read in TCP.reassembly. A segment that would zero-fill a wider gap submits the buffer and opens a new one: with no SYN the stream splits into two datagrams, with a SYN both halves report PARTIAL. Its comment picks 16 MiB from Linux's tcp_rmem and notes RFC 7323 allows windows to 2 ** 30 (a 1 GiB zero-fill); no RFC calls for the cap (fix(reassembly): TCP completion from per-buffer gaps, mod-2**32 sequence numbers, FIN waits for reordered data #1375).
• It is also a tunable outside pcapkit/utilities/config.py with no PCAPKIT_* variable (feat(utilities): one config module for the tunable limits (PCAPKIT_* environment variables), and one home per duplicated protocol constant #1726), but feat(utilities): one config module for the tunable limits (PCAPKIT_* environment variables), and one home per duplicated protocol constant #1726's inventory does not list it, so that half is weaker than point 4. On fix(reassembly): pin DTLS handshake buffers at declared message length #2469 the owner said TCP is not capped, since the RFC defines no cap.

Decision needed: does the cap stay?

1. Drop it: a segment always merges, bounded only by the sequence space.
2. Keep it as a limit in config.py (PCAPKIT_*, -1 = no limit, feat(utilities): one config module for the tunable limits (PCAPKIT_* environment variables), and one home per duplicated protocol constant #1726), default unchanged, and note on the page that it is a safety limit, not a reassembly rule.
3. Keep it as is, and record the exception on reassembly-integration.rst.
   Recommendation: 2, because the wide gap is a memory-amplification risk that config.py already handles this way (PCAPKIT_MAX_ZERO_PAD_LENGTH), and -1 gives point 4 its uncapped mode.
   Fix: on 2: add the variable, read it in TCP.reassembly, update TCP.__window__'s docs, test both sides of the cap and -1. It edits config.py (also edited by feat(protocols)!: verify and compute one's-complement checksums (#2470) #2543) and the page (also edited by refactor(foundation): route DTLS reassembly via toolkit and analysis (#2496) #2528).