feat(setup): add Hermes Agent hook integration

[Grivn]

Summary

Adds mnemon setup --target hermes using Hermes Agent's native shell hook surface instead of a MemoryProvider adapter.

• detects Hermes via hermes or ~/.hermes
• writes ~/.hermes/skills/mnemon/SKILL.md
• installs Hermes shell hooks under ~/.hermes/agent-hooks/mnemon/
• patches ~/.hermes/config.yaml for on_session_start, pre_llm_call, post_llm_call, and optional on_session_finalize
• supports eject cleanup for owned skill, hooks, config entries, and transient hook state
• documents the new target in English and Chinese usage docs

This is one native-hook implementation path for #2; it intentionally does not close the issue because a Hermes MemoryProvider/plugin path may still be worth tracking separately.

Validation

• go test ./internal/setup ./cmd
• go build -o mnemon .
• bash scripts/check_bilingual_sync.sh
• git diff --check
• isolated HOME smoke test for mnemon setup --target hermes --yes
• isolated HOME smoke test for mnemon setup --eject --target hermes --yes

Note: go test ./... still fails in existing harness UI/read tests that depend on real project lifecycle fixture data (TestPagesRenderRealData, TestLoadRealProjectRendersData); those failures are unrelated to this change.

[elessarrr]

'...it intentionally does not close the issue because a Hermes MemoryProvider/plugin path may still be worth tracking separately.' could you flesh out a bit more what you were expecting here please?

[Grivn]

Sure. This PR covers the standard Mnemon integration path: hook + skill, using Hermes' native hook and skill surfaces.

I kept #2 open because the issue also surfaced another interesting direction: a Hermes MemoryProvider / plugin integration. This is the comment I had in mind: #2 (comment)

That seems like a valid alternative for users who want Mnemon to behave more like a first-class Hermes memory backend.

I haven't looked deeply enough yet to say whether that should become an official path, and it has different tradeoffs from the hook approach, especially around automatic memory writes vs. Mnemon's LLM-supervised writeback model. So I'd rather keep the issue open for now and evaluate that separately.

[Grivn]

@elessarrr just adding an explicit mention here so this doesn't get lost: #49 (comment)

[elessarrr]

Thanks for clarifying, @Grivn. This PR covers my original proposal; glad it shipped. I’ll step back for now and keep an eye on issues in case an interesting opportunity to contribute comes up.