fix: support Codex integration requirements

[YOMXXX]

Summary

• Add host capability metadata and a standalone memory-tdai CLI entry for non-OpenClaw hosts such as Codex.
• Harden offload/standalone LLM integration: secure backend TLS defaults, custom CA support, providerOptions passthrough, multi-slash provider/model parsing, and Unicode-safe storage sanitization.
• Add L1 memory scope (global|project|session) plus independent L2/L3 stage switches, schema updates, migration compatibility, and regression tests.
• Fix gateway multi-user isolation by forwarding a stable userId to the offload backend and exposing process/data-dir diagnostics in /health.
• Add ZeroEntropy native embedding support (POST <baseUrl>/v1/models/embed) and preserve qclaw proxyUrl / embedding timeoutMs through store creation.
• Document installation into an existing Hermes Agent and add ZeroEntropy embedding examples.

Test Plan

• npm test
• npm run build
• npx tsc --noEmit --module NodeNext --moduleResolution NodeNext --target ES2022 --allowImportingTsExtensions --types node --skipLibCheck index.ts src/**/*.ts
• npm pack --dry-run --json
• git diff --check

Issue coverage

• Addresses [Feature] Support ZeroEntropy native embedding API #68 and [Feature] 希望有一篇如何集成到hermes agent的文档 #69.
• Improves diagnostics for Suggestion: improve observability for session and L1 pipeline status #64.
• Avoids implementing Feature Request: Expose tdai_memory_write tool for subagent sessions #66 direct writes in this PR because maintainers noted the bypass risk for L1 dedup/conflict detection.
• Leaves Localization issue: Hardcoded Chinese prompts in L1/L2/L3 extraction layers #38 prompt localization for a dedicated prompt/template refactor, matching maintainer feedback that it should land in a next version.

[Maxwell-Code07]

感谢你投入的精力，这个PR非常全面！我们会尽快评审相关内容。由于改动范围较大，评审、reveiw的时间可能稍长，感谢耐心等待！另外，建议后续尽量单个PR对应单个问题，能更便于高效核验迭代哈