fix: remove duplicate reward episode metrics

[TATP-233]

Summary

• Removed RewardManager._episode_sums and its per-term episode accumulation.
• Stopped emitting Episode_Reward/<term> reset extras, eliminating the duplicate TensorBoard reward tree.
• Preserved reward/<term> step metrics and stateful reward-term reset calls.
• Updated manager tests to assert that reset extras contain no episode reward metrics while reset IDs still reach stateful terms.

Linked Work

• Issue: closes Remove duplicate Episode_Reward per-term TensorBoard metrics #1569
• Parent roadmap (when applicable): not applicable
• Roadmap declared base (when applicable): not applicable
• Milestone: none
• Base branch: main

Validation

• make test-all passed on the final local head before this PR was created or updated
• Additional task-specific validation listed below

Commands actually run:

uv run pytest tests/managers/test_core_managers.py -q
uv run ruff format --check src/unilab/managers/reward_manager.py src/unilab/envs/manager_based_rl_env.py tests/managers/test_core_managers.py
uv run ruff check src/unilab/managers/reward_manager.py src/unilab/envs/manager_based_rl_env.py tests/managers/test_core_managers.py
git diff --check
make check
make test
make test-all

Remote CI route:

• Base main: all current-head checks passed for commit 8f0319c773aebbf20f17b2383c10e2c175d0c637.
• CI run: https://github.com/Motphys/UniLab/actions/runs/34813288051
• Docs build: https://github.com/Motphys/UniLab/actions/runs/34813287993

Impact

• Backend impact: none
• Platform impact: none
• Training effect expected: yes — TensorBoard per-term logs are reduced to reward/<term> only; policy input/output and optimization behavior are unchanged.

Artifacts

• W&B: none
• benchmark result: none
• video / screenshot: none
• ONNX / checkpoint: none

Checklist

• Added or updated tests where needed
• Updated docs if behavior or workflow changed
• Linked the driving issue
• Noted any follow-up work explicitly