feat: add allowedAgents config for per-agent memory isolation

[kagura-agent]

Summary

Adds an optional allowedAgents config option to restrict memory operations (recall and capture) to specific agents. Fixes #27.

Problem

In multi-agent OpenClaw setups, the supermemory plugin fires for every agent that shares the instance. Memories meant for one agent bleed into unrelated agent sessions.

Solution

New config option:

{
  "plugins": {
    "entries": {
      "openclaw-supermemory": {
        "config": {
          "allowedAgents": ["navi", "research-bot"]
        }
      }
    }
  }
}

When allowedAgents is set, both before_agent_start (recall) and agent_end (capture) hooks check ctx.agentId against the list. Non-matching agents are silently skipped with a debug log.

When allowedAgents is omitted or empty, all agents are allowed (preserving current behavior — no breaking changes).

Changes

• config.ts: Added allowedAgents field to config type, parsing, JSON schema, and isAgentAllowed() helper
• hooks/recall.ts: Guard clause using isAgentAllowed()
• hooks/capture.ts: Guard clause using isAgentAllowed()

3 files changed, +34 lines.

Notes

• Uses ctx.agentId (not ctx.sessionKey as suggested in Feature request: per-agent memory isolation via allowedAgents config #27) since agentId is the stable identifier for which agent is running
• Follows the existing SKIPPED_PROVIDERS pattern in capture.ts for guard precedent
• isAgentAllowed() is exported from config.ts for potential reuse in tools/commands if needed later