fix: make dev_reload work for files in nodes/

[keturn]

Summary

The dev_reload option to load Python changes on the fly should work for custom nodes as well as Invoke's own sources.

(Have y'all been trying to write nodes without this? Or do you have another hot-reloading mechanism you've been using instead?)

QA Instructions

1. have some code in a nodes/ subdirectory.
2. Run invoke with INVOKEAI_DEV_RELOAD=true environment variable. (Or set dev_reload = True in your `invokeai.yaml'.)
3. Run invokeai-web
4. Startup logs should include a [jurigged] Watch line for your nodes source.
5. When you edit that file, the log should include a [jurigged] Update line for it.

Merge Plan

N/A

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• N/A Tests added / updated (if applicable)
• N/A Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

[keturn]

Alternatively, we could make load_custom_nodes explicitly communicate with jurigged, but this is easier and almost as effective.

The one gap is if someday in the future we change it so load_custom_nodes is called again later somewhere else.

[psychedelicious]

Nice, thanks.

I expect this to work only for internal logic changes to the nodes.

For example, the logic within the invoke() method.

I don't think it will work to add/remove nodes or change node input/output signatures.

Calling BaseInvocation.invalidate_typeadapter() after the hot-reload should fix adding/removing nodes and changed node input signatures.

In #7826, invocation registration is updated and a new method to invalidate the output classes is added. So calling InvocationRegistry.invalidate_invocation_typeadapter() and InvocationRegistry.invalidate_output_typeadapter() should fully enable hot-reload for nodes.

[psychedelicious]

Merging this as-is now to unblock anything you may be done - we can get the full hot reload working in the future.