feat(hermes): add offload context engine

[RerankerGuo]

Description | 描述

Add a standalone Hermes ContextEngine plugin for TencentDB short-term offload under hermes-plugin/context_engine/tencentdb_offload.

This is a narrow runtime follow-up to the Hermes short-term memory gap:

• Adds TencentdbOffloadContextEngine with the expected name, should_compress, compress, and update_from_response methods.
• Keeps the implementation HTTP-only so it does not import TencentDB TypeScript internals or the existing Hermes memory provider.
• Sends remote compaction requests to POST /v2/offload/compact and accepts either messages or compacted_messages response shapes.
• Sends best-effort ingest payloads to POST /v2/offload/ingest on a daemon thread so the user path is not blocked.
• Provides a local tail-truncation fallback that preserves system messages and recent conversation context if the remote service is down or returns an unsupported response.
• Documents environment variables and the expected remote contract.
• Adds stdlib unit tests for threshold decisions, remote compaction, fallback behavior, and async ingest scheduling.

I intentionally did not change the TypeScript Gateway routes in this PR. The plugin defines the Hermes-side adapter and remote HTTP contract first; Gateway-side /v2/offload/* implementation can remain a follow-up if maintainers want it in this repository rather than an external offload service.

Related Issue | 关联 Issue

Related to #102

Change Type | 修改类型

• Bug fix | Bug 修复
• New feature | 新功能
• Documentation update | 文档更新
• Code optimization | 代码优化

Self-test Checklist | 自测清单

• Verified locally | 本地验证通过
• No existing features affected | 无影响现有功能

Verification | 验证

• python3 hermes-plugin/context_engine/tencentdb_offload/tests/test_context_engine.py
• python3 -m py_compile hermes-plugin/context_engine/tencentdb_offload/__init__.py hermes-plugin/context_engine/tencentdb_offload/client.py hermes-plugin/context_engine/tencentdb_offload/fallback.py hermes-plugin/context_engine/tencentdb_offload/tests/test_context_engine.py
• npm test
• npm run build
• git diff --check

Additional Notes | 其他说明

The scope is deliberately conservative: it makes the Hermes ContextEngine slot usable and failure-tolerant without changing the existing memory_tencentdb provider or the OpenClaw/offload TypeScript runtime.

[Maxwell-Code07]

Thanks for your attention! We’ll review this and get back to you as soon as possible.