#10676: Claude/authorize external access lbcnb

<!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> - Adds a new “Aether Conductor” module that wraps an agent process, analyzes output for external-access requests, forwards them for operator authorization, executes browser actions, and injects results back into the agent. - Introduces a conductor CLI (`openclaw conductor ...`) plus config types and wiring into the gateway RPC layer (`conductor.request/resolve/status/history`). - Adds a regex/Gemini/OpenAI analyzer implementation with tests and integrates messaging-based authorization forwarding. - Adds an agent tool (`conductor`) to explicitly request external access through the gateway. <h3>Confidence Score: 2/5</h3> - This PR adds substantial new behavior but has a few correctness and safety issues that should be fixed before merging. - Score reduced due to a real command-injection risk from spawning wrapped commands with `shell: true`, plus several concrete correctness issues (duplicate stop events, config option mismatch, missing resolver identity in history, and unused imports that may break CI depending on lint settings). - src/conductor/interceptor.ts, src/conductor/conductor.ts, src/conductor/browser-executor.ts, src/gateway/server-methods/conductor.ts <!-- greptile_other_comments_section --> _{(3/5) Reply to the agent's comments like "Can you suggest a fix for this @greptileai?" or ask follow-up questions!} <!-- /greptile_comment -->