refactor: share agent runtime hygiene helpers

[vaskarvelas]

Summary

Extract shared runtime-hygiene helpers for the first-party agent stores and route both admin and onscreen lifecycle message processing through the same helper layer.

TL;DR;
It helps the agent save tokens even more.

Why

Both stores were carrying overlapping logic for:

• cloning conversation messages
• sanitizing assistant-message evaluation input
• applying processed message mutations safely
• routing processed lifecycle messages through the existing extension seams

That duplication made it easier for the two runtimes to drift and harder to add new lifecycle processing consistently.

What changed

• add _core/agent-chat/runtime-hygiene.js
• move shared message clone/apply/normalize/resolve/create helpers there
• update admin store to use shared helpers for:
  • submit
  • assistant-response
  • history-compact
  • protocol-retry
  • execution-output
• update onscreen store to use the same shared helper flow
• keep existing processAdminAgentMessage / processOnscreenAgentMessage seams intact
• add tests/runtime_hygiene_test.mjs
• update owning docs and supplemental runtime docs for the new shared helper contract

Safety / behavior notes

This is intended as a refactor and seam consolidation, not a product behavior change. The existing extension seams remain in place; the new helper layer only centralizes shared runtime hygiene and mutation isolation.

Test plan

Passed:

• node --test tests/runtime_hygiene_test.mjs tests/assistant_message_evaluation_test.mjs
• node --check app/L0/_all/mod/_core/admin/views/agent/store.js
• node --check app/L0/_all/mod/_core/onscreen_agent/store.js
• node --check app/L0/_all/mod/_core/agent-chat/runtime-hygiene.js