tests: make the Docker integration suite build and boot again

[TeoSlayer]

Summary

The compose-based suite in tests/integration/local has not been able to start since rendezvous and gateway left this repository (a6f97f5, 2026-05-27). Each test brought the stack up with output discarded, polled for 60 s and failed at "agents did not register", so every run reported setup failures and nothing about the product.

With this change the image builds, the stack boots in about 7 s, and tests run. On main (macOS, 8 workers, 12 minutes): 150 tests, 109 pass, 41 fail.

What was broken

Each problem hid the next one.

#	Problem	Change
1	Dockerfile.multi built ./cmd/rendezvous and ./cmd/gateway, which no longer exist	Build the rendezvous server from the module version go.mod pins. Drop the gateway build
2	No main package for pilot-gateway exists anywhere (the gateway repo is a library and is archived)	Remove the ten test_gateway_* scripts and their compose file
3	The registry's /api/stats now requires an admin token, so the rendezvous health check never passed and agents never started	All 217 uses only read total_nodes; point them at the unauthenticated /api/public-stats
4	macOS has no timeout(1); 38 scripts call it on the host and reported "command not found" as a product failure	run-all.sh puts shims/timeout on PATH when the command is missing
5	run-all.sh ran test_cli.sh on the host. That script is the in-container driver: it starts a daemon on /tmp/pilot.sock, and its cleanup runs pkill -9 -f pilot-daemon and deletes that socket, so a full run killed the developer's own daemon	Exclude it; _lib.sh no longer runs pkill on the host
6	QUICK.txt listed 17 scripts that do not exist	Pruned

The 41 remaining failures

These are not addressed here. They are what the suite reports now that it runs.

Cause	Count	Tests
pilotctl send-file output no longer has .data.ack	12	send_file_* (4), size_file_* (4), mixed_traffic_burst, stress_edge, rendezvous_restart_midflight, sender_sigkill_midfile
pilotctl inbox lists preview, not data	7	message_text, message_json, message_binary, message_payload_integrity, chain_abc_message, mesh_3agents_crosstraffic, inbox_ordering
Relay fallback with the direct path blocked	3	force_relay_send_message, force_relay_send_file, force_relay_pubsub. A real daemon problem: see #505. The scripts also send 10 s after the partition, before the daemon has moved to the relay
Compose stack did not come up	2	nat_asymmetric_routing, nat_dual_symmetric
Need individual triage (CLI, registry or policy behaviour changed since May)	17	beacon_ping, broadcast, nat_egress_443_only, nat_hairpin, nat_partition_post_reg, nat_stateful_firewall, net_join_deny_per_network, policy_connect_tag, policy_cycle_evict, policy_datagram_score, register_identity_new_endpoint, rendezvous_beacon_split, resilience, sec_beacon_amplification, service_agent, trust_revoke, type_routed_service

Not covered

• Dockerfile (the CLI + Python SDK image behind make test, test-cli, test-sdk, build) still builds sdk/cgo, sdk/python and ./cmd/gateway, none of which are in this repository. Those make targets remain broken.
• tests/integration/*.sh holds 48 older copies of scripts that also exist under local/. Untouched.
• Tests still discard the output of compose up, so the next build break will again look like a registration timeout.

Test Plan

• docker build -f tests/integration/local/Dockerfile.multi succeeds; the image contains pilot-rendezvous, pilot-daemon, pilotctl
• ./run-all.sh -j 8 over all 150 scripts: 90 pass before the timeout shim, 109 with it
• ./run-all.sh --list no longer includes test_cli.sh or any gateway script
• A full run leaves a daemon running on the host untouched
• Not run on Linux

🤖 Generated with Claude Code