Fix sparse Wipe marker rewards

[tandede]

What this does

Fixes #418.

Wipe.reward() documents unit_wiped_reward as a discrete reward for every marker wiped during the current step. The normalization factor also includes num_markers * unit_wiped_reward, so that component is part of the intended return for both reward modes. However, the implementation added it only when reward_shaping=True. With sparse rewards, newly wiped markers were made transparent and recorded in wiped_markers, but contributed no reward unless the final marker also completed the task.

This change moves only the per-marker increment outside the dense-reward condition. Dense rewards keep their existing distance, contact, force, acceleration, and completion components; sparse rewards now receive the documented marker reward while retaining the existing collision and completion behavior.

A focused regression invokes the reward calculation with a minimal simulated Wipe state and verifies that, with reward_shaping=False, a newly wiped marker:

• contributes exactly one unit_wiped_reward;
• is recorded in wiped_markers; and
• has its visual geometry hidden.

How it was tested

• python -m pytest tests/test_environments/test_wipe_reward.py tests/test_environments (5 passed)
• python -m pytest (262 passed, 61 skipped)
• pre-commit run --files robosuite/environments/manipulation/wipe.py tests/test_environments/test_wipe_reward.py
• The two flake8 commands used by .github/workflows/run-tests.yaml (strict error count: 0)
• git diff --check

How to checkout & try? (for the reviewer)

python -m pytest tests/test_environments/test_wipe_reward.py