Dynamic Model Discovery for LLM Nodes

[asclearuc]

LLM provider nodes currently hard-code their model lists inside services.json
files. When providers release new models or deprecate old ones, a developer must
manually edit the JSON — a fragile, error-prone process that causes model lists
to fall behind the real provider catalogue.

Create a sync tool that discovers currently-available models directly from provider APIs and propagates
that information into the services configuration. The tool should support all cloud LLM providers that are currently present: OpenAI, Anthropic, Gemini, Mistral, DeepSeek, xAI, Perplexity, and Qwen.

Key behaviours:

• Dry-run by default. Changes are previewed before being written. --apply
  commits them to disk.
• Non-destructive. Existing profile titles, model fields, and manual customisations are preserved.
• Deprecation tracking. Models no longer listed by a provider are marked
  deprecated: true with a migration message, rather than silently removed.
  Existing pipelines continue to work.
• Token limit accuracy. Limits are resolved automatically.
• CI/CD ready. Designed to run on a weekly schedule, opening a pull request
  with the full add/update/deprecate summary for developer review before merge.
• If a key for the corresponding LLM provider is present, the tool runs a smoke test for that provider.

[github-actions]