test: cover tcp_conntrack seq wraparound + exhaustive FSM index scan

[w180112]

Summary

Two residual tcp_conntrack gaps from the test-coverage analysis of the PR #72/#75 conntrack work: the seq/ack window tests only exercised deltas far from the 32-bit wrap point, and the constructor-built O(1) FSM dispatch index was never verified against the full state × event space. Test-only change; no production code touched.

New cases in unit_test/pppd/tcp_conntrack_test.c (19 assertions)

test_seq_wraparound (8)

• tcp_conntrack_seq_update advances max_seq_end/max_ack across the 0xFFFFFFFF → 0 boundary (wrap-aware signed compare: unsigned distance ~4G, signed delta tiny positive)
• Monotonicity across the wrap: once the baseline crosses 0, a stale pre-wrap seq/ack (huge unsigned value, negative signed delta) cannot roll the max back
• The explicit == 0 seed guard fires for an ISN ≥ 2³¹ — without it the ISN looks "older than zero" under signed math and the baseline would never advance
• tcp_conntrack_seq_valid accepts in-window seq and ack just past the wrap, and still enforces the 16 MB seq slack and 0xFFFF ack slack across the wrap (both drop sides verified)

test_fsm_index_full_scan (11)

• Mirrors tcp_conntrack_tbl as an expected-next-state matrix and drives tcp_conntrack_fsm through all 10 × 6 (state, event) combinations:
  • every defined row must land on the table's next_state
  • every undefined combo must keep the current state and return SUCCESS (no state may hit the no-rows ERROR contract, since every state has at least one row)
• One assertion per state (with first-mismatch detail) plus a return-value sweep — future table edits or constructor index-build drift get caught here immediately
• ACK rows are driven LAN-side; the WAN-side MID_STREAM → ESTABLISHED promotion is handler-driven and already covered by the existing promotion tests

Existing cases untouched.

Testing

• make test: 1404/1404 pass
• e2e_test/run_e2e_test.sh: 73 PASS / 0 FAIL / 0 SKIP