[BUG] Responses chaining ignores agent session ids when Langfuse is disabled

## Summary When CAMEL runs response-mode backends without Langfuse enabled, response chaining falls back to the shared `"__default__"` session key even when `ChatAgent` has already set an agent session id. This means two different agent sessions can reuse the same `previous_response_id` if they share one backend instance. ## Why this is a real bug CAMEL is a stateful multi-agent framework. Session isolation should not depend on whether Langfuse tracing happens to be enabled. Today: - `ChatAgent` calls `set_current_agent_session_id(self.agent_id)` - but `get_current_agent_session_id()` returns `None` unless Langfuse is configured - so `OpenAIModel` / `OpenAICompatibleModel` collapse to `"__default__"` That makes response chaining reuse state across unrelated agent sessions. ## Affected code - `camel/utils/langfuse.py` - `camel/models/openai_model.py` - `camel/models/openai_compatible_model.py` ## Minimal reproduction I validated this locally on current `master` by: 1. creating one response-mode backend instance, 2. saving chain state under agent A, 3. switching to agent B, 4. calling `_prepare_responses_input_and_chain(...)`. Observed result: - session key stayed effectively shared, - `previous_response_id` from agent A was reused for agent B. ## Expected behavior Agent/session scoping for response chaining should work even when Langfuse is disabled. ## Actual behavior Session scoping currently depends on Langfuse availability, which causes cross-session state bleed in normal non-Langfuse deployments. ## Suggested fix direction Decouple `get_current_agent_session_id()` from Langfuse tracing availability, or otherwise keep a separate always-on context-local session key for model backends.