refactor: move current time from system prompt to user message for prompt cache stability

[wuyuanfr]

Summary

Move the current-time injection from the system prompt to the user message side, so the rendered system prompt becomes fully deterministic per agent config version — enabling prompt/KV cache hits across requests.

Changes

File	Change
backend/prompts/manager_system_prompt_template_{en,zh}.yaml	Remove it is {{time}} now / 现在是{{time}}
backend/prompts/managed_system_prompt_template_{en,zh}.yaml	Same as above
backend/utils/context_utils.py	Drop time_str param and "Current time" line from _format_app_context, build_skeleton_header_component, build_context_components, build_app_context_string; remove unused datetime import
backend/agents/create_agent_info.py	Remove time from render_kwargs and build_context_components call; remove datetime import
sdk/nexent/core/agents/core_agent.py	In run(), prepend [Current time: …] to self.task so the timestamp travels with the user message
test/backend/utils/test_context_utils.py	Drop time_str= from 3 test cases

Why

A timestamp baked into the system prompt changes on every request, preventing the LLM prompt/KV cache from matching across requests. By shifting the time to the user-message prefix, the system prefix becomes static text — the rendered system prompt is now identical for any given (agent_id, tenant_id, version_no, language) tuple, making it directly cacheable.

Test

All 5 related test suites pass individually (389 cases total).

Notes

• The EN templates never had user_id; this PR only removes time and does not add user_id to EN, preserving the existing EN/ZH asymmetry (can be addressed in a follow-up PR).
• skill_creation_agent uses an independent prompt and is unaffected.
• Side benefit: prompt_template["system_prompt"] Jinja rendering is now fully deterministic, enabling a (agent_id, tenant_id, version_no, language)-level cache in create_agent_config for further CPU savings.