What does this PR do?

Adds per-platform Honcho user-peer identity routing for gateway contexts.

This is related to #15162 and the Honcho consolidation PR #15381. Those changes add pinPeerName, which is a good opt-in fix for the common personal deployment case where every gateway should resolve to the configured peerName.

This PR covers the next step: mixed routing. A personal Hermes deployment may want Telegram and WhatsApp to share one Honcho peer, while keeping WeChat or Discord scoped to the platform-native user_id.

Fix

Adds two config fields:

• peerIdentityStrategy: global default strategy, preserving existing runtime behavior by default.
• peerIdentities: per-platform override map.

Supported rule values:

• runtime: use gateway user_id, falling back to peerName.
• configured: use peerName, falling back to gateway user_id.
• any other string: use that string as a custom Honcho peer name for the platform.

Example:

{
  "peerName": "kawarox",
  "peerIdentityStrategy": "configured",
  "peerIdentities": {
    "telegram": "configured",
    "whatsapp": "configured",
    "weixin": "runtime",
    "matrix": "matrix-custom-peer"
  }
}

In that setup:

The resolved peer is passed into HonchoSessionManager, so Honcho writes, context injection, dialectic calls, and conclusion/profile operations use the same user peer.

Files changed

• plugins/memory/honcho/client.py
  • Adds peer_identity_strategy and peer_identities to HonchoClientConfig.
  • Adds resolve_runtime_peer_name() to resolve global and per-platform rules.
  • Parses peerIdentityStrategy and peerIdentities from root or host config.
• plugins/memory/honcho/__init__.py
  • Resolves the runtime Honcho user peer before creating HonchoSessionManager.
• plugins/memory/honcho/README.md
  • Documents the new config keys and examples.
• tests/agent/test_memory_user_id.py
  • Adds coverage for default runtime behavior, configured-peer routing, and per-platform custom routing.

Relationship to pinPeerName

If maintainers prefer to keep pinPeerName as the public API for the all-platform case, this PR can be adjusted to layer per-platform routing on top of it instead of introducing a separate global strategy name:

• pinPeerName: true remains the simple global "configured peer wins" mode.
• peerIdentities only handles per-platform overrides for mixed gateway deployments.

I left the branch as a draft because #15162/#15381 may land first, and this PR may need a small rebase or API naming adjustment afterward.

Related issues / PRs

• Related: honcho: peerName in config should win over platform user ID for single-user setups #14984
• Related: Honcho memory plugin ignores gateway user_id when peer_name is configured #14401
• Related: fix(honcho): pinPeerName opt-in keeps memory unified across platforms (#14984) #15162
• Related: fix(honcho): bug-fix consolidation — 7 upstream PRs, preserved authorship #15381

Type of Change

• Feature (non-breaking opt-in configuration)
• Tests (adding or improving test coverage)
• Documentation

Test plan

• venv/bin/python -m pytest tests/agent/test_memory_user_id.py -q

Result: 16 passed.

Not in scope

• Migrating existing conclusions/messages between Honcho peers.
• Removing old runtime peers from existing Honcho sessions.
• Changing the default multi-user gateway behavior.