terminated和done

按照gym.Env.step接口定义:

- terminated表示 s_t 是MDP终止状态（智能体成功、失败、死亡），s_t+1 不存在，计算TD-Error时不能再使用 Q_t+1 或 V_t+1

- truncated表示仿真被截断（达到最大仿真步或仿真异常），s_t+1 可能存在

- done = terminated or truncated，done时 s_t+1 也可能存在

但最新版roll/agentic/env中的几个环境实现直接令 terminated = done，但这两个参数含义通常情况下不等价。

另外，希望增加类似register(id, config_class, env_class)接口，方便将自定义环境注册到REGISTERED_ENVS和REGISTERED_ENV_CONFIGS中

Read the gym.Env.step contract first, then inspect the environment implementations under roll/agentic/env and the REGISTERED_ENVS and REGISTERED_ENV_CONFIGS structures. Done means terminated and truncated preserve their distinct meanings while done is their OR, and custom environment/configuration classes can be registered through a suitable interface.

Written by the indexing model from the issue text.