net-combat-events: pick the smoker through PickWeapon, so tier:ci finds it

[Laeresh]

The engine suites job (tier:ci, headless) has been red since the review-fixes merge (a08972a). net-combat-events fails the check "the smoker resolves and the host's seat carries a pylon and the session's screens"; the other 195 suites pass.

SmokeEvents looked the smoker up as wep_13, which only a real extraction holds. CI runs on the --synthetic-data tree, whose weapons.json carries the smoker as wep_probe_smoker, so the lookup failed before the check ran. The merge was verified on an install, where wep_13 exists.

SmokeEvents now goes through OrdnanceSuites.PickWeapon, as the ordnance smoke suites do: wep_13 on an extraction, the first SMOKE_SCREEN record on the synthetic tree. ordnance-launch-axis, already on tier:ci, launches that synthetic record from swapped pylons the same way.

Not run locally (no dotnet, pwsh or Godot in the authoring environment): the build, RunTests.ps1 and the engine suites. This PR's CI run is the check.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Gwi3N7pb5jKPQ7rXa5wbXW

---

Generated by Claude Code

[Laeresh]

CI is green on c535b73 (attempt 2 of run 37938065755).

Attempt 1's engine suites job passed all 196 tier:ci suites, net-combat-events included, but Godot then exited 139 in its teardown after writing the report, which RunCiSuites.ps1 counts as a failure. The re-run of that job on the same commit passed cleanly. On the synthetic tree the newly reachable SmokeEvents path builds no smoke emitters (the fixtures carry no generate_smokescreen), so it adds no nodes to tear down. The crash may be the Godot 4.7 dummy-renderer thread race the CI allowlist already names (godotengine/godot#121949). The crash's godot.err was not read, because the artifact download was blocked from the session that wrote this fix. If an exit 139 recurs, that file is where to look.

---

Generated by Claude Code