hermes - 💡(How to fix) Fix Detect duplicate gateway supervisors in doctor/status diagnostics

Problem

A profile can be supervised by more than one systemd lane at once, e.g. a system service under /etc/systemd/system/ plus a stale user service under ~/.config/systemd/user/. When both run gateway run --replace, they intentionally kill/replace each other and look like a gateway crash loop.

In non-login shells, plain systemctl --user can fail with Failed to connect to bus, so operators may falsely conclude there are no user services.

Desired outcome

Hermes diagnostics should make duplicate gateway supervisors obvious before users chase provider warnings, MCP sockets, or unrelated traceback noise.

Suggested behavior

Add duplicate-supervisor detection to one or more of:

• hermes doctor
• hermes gateway status
• hermes gateway install/start/restart preflight

Detection should inspect, where available:

• system unit files/states: systemctl list-units/list-unit-files 'hermes-gateway*'
• user unit files/states with a usable bus:
  • XDG_RUNTIME_DIR=/run/user/$(id -u)
  • DBUS_SESSION_BUS_ADDRESS=unix:path=$XDG_RUNTIME_DIR/bus
  • systemctl --user list-units/list-unit-files 'hermes-gateway*'
• live gateway process commands and scoped HERMES_HOME/profile PID files

Acceptance criteria

• A duplicate user+system service for the same profile is reported with the conflicting unit names and profile/HERMES_HOME.
• The diagnostic distinguishes "could not connect to user bus" from "no user services exist".
• The message recommends choosing one supervisor instead of editing model/provider config.
• Safe docs or remediation hints are included, without blindly deleting unit files.