Quaternion deviation logging, MAVLink query tooling, IC-capture fix, venv centralization

[mcroomp]

Summary

• calibrate run: log actual vs. target attitude quaternion (mav_att_q_*, mav_att_target_q_*, mav_att_qerr_*) and show qerr(d)/qyaw(d) in the live console table instead of V/A.
• groundstation/gcs.py: fix ATTITUDE_TARGET (#83) decode-registration bug (was only registered under SET_ATTITUDE_TARGET); explicitly request ATTITUDE_QUATERNION/ATTITUDE_TARGET via MAV_CMD_SET_MESSAGE_INTERVAL since neither rides the legacy EXTRA1 stream group.
• scripts/rawes.lua: fix a real MODE_PASSIVE --hold (RAWES_YIC sentinel) IC-capture bug — _ic_seeded persists across RAWES_MODE transitions for the whole FC boot, so a second --hold capture in the same boot only updated the pending fields and never reached the committed GUIDED target, leaving the attitude target frozen at a stale prior value (large non-decaying quaternion error). Now commits directly to _ic_roll_deg/_ic_pitch_deg in the capture branch.
• analysis/mavlink_jsonl_query.py: reassemble multi-chunk STATUSTEXT messages (ArduPilot splits >50-byte gcs:send_text() calls across messages sharing an id/chunk_seq) instead of printing raw ~50-byte fragments.
• analysis/mavlink_jsonl_query.md: new canonical usage doc for the tool.
• setup.cmd/setup.sh: always perform the editable pip install -e . install (previously missing entirely, causing ModuleNotFoundError in fresh venvs); now hash-gated on pyproject.toml like the requirements.txt install so it's a no-op after the first run.
• calibrate.cmd/pump_lua.cmd/visualize.cmd: centralize venv setup by delegating to setup.cmd instead of duplicating setup logic.
• Docs: design/flight_stack.md, design/calibration.md, design/sitl_testing.md, AGENTS.md updated to document the above (RAWES_YIC capture path + gotcha, quaternion columns, mavlink_jsonl_query.py as first-line diagnostic tool).
• .gitignore: fix tests/oneoff/ rule to **/tests/oneoff/ so it also matches the stray simulation/tests/oneoff/ legacy directory (was leaving 46 untracked scratch files visible to git status).

Testing

• Verified pip install -e . --no-deps idempotency and hash-gating via bash setup.sh (first run installs + stamps, second run skips both installs).
• Verified analysis/mavlink_jsonl_query.py statustext reassembly against run_passive_20260722_121402.mavlink.jsonl and an older run_passive_20260715_105728 log.
• Diagnosed the --hold staleness bug end-to-end via the mavlink log (RAWES YIC capture: r=89.9 p=20.0 vs. subsequent RAWES guided cmd: ... rpy=(0.0,0.0,...) never converging).
• Full SITL/simtest re-run of the Lua fix was not performed in this session (would require spinning up SITL).