stacks: commit full_default + lite_default wiring baselines for MIGHTY (0.21.0-dev.12)

[andrewjong]

Summary

Commits the two wiring baselines that #424 left to the bootstrap path, from validated Isaac Sim snapshot runs (airstack test -m wiring --stack <name> --sim isaacsim --num-robots 1) on the 0.21.0-dev.11 robot image built from the current Dockerfile.robot:

Stack	Nodes / edges	vs the previous DROAN baseline (node set)
full_default	86 / 392	mighty/{global_mapper_ros,mighty_node,mighty_bridge,mighty_init_pose_tf} in, droan/disparity_expander_node out
lite_default	83 / 378	same swap; differs from full_default only by Container, vdb_mapping, random_walk_node

The READMEs point at the committed files again. VERSION → 0.21.0-dev.12 (docs/baselines only; publish retags).

Heads-up: MAVROS 2.15.1 renamed plugin sub-nodes

The fresh image carries MAVROS 2.15.1 (the dev.10 image had 2.14.0). Plugin sub-nodes are now interface/mavros/mavros/<plugin> instead of interface/mavros/<plugin>. Both new goldens were captured on the new image, so they match what CI will observe on a fresh build. The pre-existing goldens (full_droan, full_droan_cpu, full_macvo) were captured on 2.14.0 and will report drift on freshly built images until regenerated (airstack test -m wiring --stack <name>); that is a trunk-wide follow-up, not caused by this PR, and is recorded in the release notes. A first capture of full_default on an interim dev.10-based image was discarded for exactly this reason — after normalising the MAVROS names it was node-identical to the committed one.

Validation

• pytest tests -m wiring --stack full_default and --stack lite_default: sentinel check passed, observed snapshots written; full_default re-run against its own new golden is the natural next CI check.
• Stack layout contract (tests/meta/test_stack_layout_contract.py): 42 passed (H1 names the stack, wiring-graph-v1 trailer parses).
• Operational note for whoever regenerates next: a fresh robot container compiles the workspace from scratch (~13 min, the MIGHTY solver dominates), which exceeds the sentinel wait (300 s); bring the stack up once with airstack up --stack <name>, wait for the nodes, then run the wiring mark. A stale colcon tree from a different image fails with No rule to make target libfastcdr.so.2.2.x — wipe robot/ros_ws/{build,install,log} (root-owned; use a container) first.

🤖 Generated with Claude Code

[github-actions]

Test Metrics — cbeb6b331c56759a6a74a5a839de5776a5d83bf9

This was a unit/build-only run. Simulation regression comparison does not apply.

system.test_build_packages

Pass rates

Test	Pass	Fail	Skip	Rate
test_colcon_build_gcs	0	1	0	0%
test_colcon_build_ms_airsim	0	1	0	0%
test_colcon_build_robot	0	1	0	0%
test_colcon_test_robot	0	1	0	0%

Metrics

Test	Metric	Value
test_colcon_build_robot	duration_s	0.79s
test_colcon_test_robot	duration_s	0.644s
test_colcon_build_gcs	duration_s	0.606s
test_colcon_build_ms_airsim	duration_s	0.581s

[andrewjong]

Note: three commits pushed to this branch after the merge (asm_mighty v0.1.5 re-pin, refreshed baselines, and the CI fix for airstack up in a venv) never landed here; they are in #428. The description above was restored to what merged.