Add unit tests for ppo_continuous_action GAE computation

[YangNuoCheng]

Description

Towards #252, and a follow-up to the GAE unit tests for ppo.py. Adds logic unit tests for the GAE computation in cleanrl/ppo_continuous_action.py.

• test_ppo_continuous_action_gae_matches_explicit_sum — cross-checks the reverse in-place recurrence in ppo_continuous_action.py against an independent explicit discounted-sum reference A_t = Σ_{k≥t} (γλ·nnt)^… · δ_k, on randomized rollouts (with terminations). Same precedent as tests/test_jax_compute_gae.py.
• test_ppo_continuous_action_gae_terminal_masking — a hand-computed example that specifically exercises the episode-termination masking path: a done must zero out nextnonterminal and stop advantage propagation across the boundary. Expected advantages [1.25, 1.0, 1.0] are reviewer-verifiable by hand.

This targets the issue's goal of verifying logic correctness rather than black-box outputs, and the terminal-masking case adds coverage for the boundary behavior that is easy to break silently.

Types of changes

• Bug fix
• New feature
• New tests
• Documentation

Checklist

• New tests pass locally: pytest tests/test_ppo_continuous_action_gae.py (2 passed)
• No changes to existing files; only a new test module is added
• Lint-clean (black --line-length=127, isort --profile=black)

[vercel]

@YangNuoCheng is attempting to deploy a commit to the Costa Huang's projects Team on Vercel.

A member of the Team first needs to authorize it.