feat(checker): add BodyContactChecker matching Hub usage

[keyneszeng]

Summary

Adds a new success-condition checker that detects when two MuJoCo bodies are in (or near) contact. This matches the BodyContactChecker type already used in production Hub task JSON (e.g. task 4694, whose checker_config is published via the public Hub API).

Why

The Hub backend, AxisDataCleaning (Python), and this AxisWebInfra (JS) MUST agree on what counts as a successful task attempt. Until now, the JS side was missing 3 checkers that Hub already uses in production.

This PR is part 1 of 3 to restore JS / Python / Hub parity.

Changes

• New file src/components/mujoco-framework-next/checkers/BodyContactChecker.js (~270 lines incl. docstring)
• Modified src/components/mujoco-framework-next/checkers/index.js — registered the new class and added the body_contact alias

Task config schema (matches Hub)

{
  "type": "BodyContactChecker",
  "body_a": "akita_black_bowl_1_main",
  "body_b": "plate_1_main",
  "max_contact_distance": 0.02,    // meters; default 0
  "min_contact_count": 1            // default 1
}

Cross-repo coordination

• JS (this PR): walks MuJoCo's data.contact array and counts pairs (geom1, geom2) where one geom belongs to body_a and the other to body_b, with dist <= max_contact_distance.
• Python (companion PR to AxisDataCleaning): same semantics, but using AABB intersection + inflation (offline validation, no live MuJoCo).
• Hub backend: already present, uses the same field names.

Verification

• JS: manually instantiated through a real CheckerManager with mock data — constructor succeeds, check() returns the expected boolean, getStatus() produces the expected progress object.
• Python: 7/7 unit tests pass.

JS testing in real mujoco-js requires Node + mujoco-js (I don't have it locally); happy to add a vitest harness if maintainers want.

Checklist

• Follows existing checker pattern
• Options support both snake_case and camelCase (Hub uses both)
• getStatus() / getProgress() / reset() / updateModelData() implemented
• Companion Python PR follows the same algorithm