:bug: Unavailable external plugins lead to adapter crashloop

[evaline-ju]

Describe the bug

Currently if an external plugin is provided in the plugin config but not available/reachable, this will cause the plugin manager and thus the plugin adapter to crash.

Steps To Reproduce

1. Update the plugin config to utilize an external plugin such as

plugins:
  - name: NemoCheck
    kind: external
    hooks: ["tool_pre_invoke", "tool_post_invoke"]
    mode: enforce
    mcp:
      proto: STREAMABLEHTTP
      url: http://nemocheck-plugin-service:8000/mcp

2. Make the plugin service unavailable (delete service or just not deploy it yet)
3. See a log such as the following in the crashing adapter

RuntimeError: Plugin initialization failed: NemoCheck - External plugin 'NemoCheck' connection failed after 3 attempts: http://nemocheck-plugin-service:8000/mcp is not reachable. Please ensure the MCP server is running.

Expected Behavior

Provide a warning for unavailable external plugins but do not crash the adapter based on unavailability

Additional Context

No response

[julianstephen]

Potentially dependent on IBM/mcp-context-forge#2796

[evaline-ju]

Potentially dependent on IBM/mcp-context-forge#2796

I have also noticed that even when the external plugin is available but updated i.e. the deployment rolls over, I end up having to roll over the adapter as well to 're-connect'. This seems in line with the buggy behavior described

[julianstephen]

IBM/mcp-context-forge#2796 is now closed. Not a priority for first release, but can be retested with latest version.