Publish reproducible offline mjlab runtime

Replace the unavailable MuJoCo development-wheel environment with the verified Python 3.12 offline wheelhouse and update the task brief. Archive SHA-256: 18266648115105e2cc4b3366fea0a6d52dcedeee4db63723e3f0298dfe5cda3c.

tasks/engineering/humanoid_wbc_policy_evaluation/base/input/runtime_env/mjlab.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:23f4d52df7fea3dcc33ddd55fab38d084704eea99c59e702203b9de88650ef0b
-size 217733393
+oid sha256:18266648115105e2cc4b3366fea0a6d52dcedeee4db63723e3f0298dfe5cda3c
+size 758586982

tasks/engineering/humanoid_wbc_policy_evaluation/base/input/task_brief.md

@@ -6,6 +6,12 @@ provided locally in `input/runtime_env/motions-1.zip` and
 `input/runtime_env/policies.zip`; do not log in to W&B and do not fetch private
 W&B artifacts.
 
+The mjlab archive is self-contained for Linux x86_64. After extracting it, run
+`bash install_offline.sh ./mjlab-runtime`. This verifies the bundled CPython,
+`uv`, requirements hashes, and wheelhouse without network access. Use
+`./mjlab-runtime/bin/play` in place of the leading `uv run play` tokens in each
+listed command; keep every remaining argument unchanged.
+
 For each case, unpack the mjlab runtime archive and the two asset zips into your
 working directory, run or inspect the listed `play_command`, watch the policy
 behavior against the reference motion, save a visible motion demo, and classify