Fix AI service lint regression in CI by restoring top-level import order and lazy LiteLLM call path

[Copilot]

CI lint was failing in server/src/flowforge_server/services/ai.py due to a non-top-level import (E402) and a direct litellm reference that bypassed the file’s lazy-loader pattern. This PR applies a minimal correction to align the module with existing import/lazy-load conventions.

• Import ordering cleanup

  • Moves get_settings into the module import block so all imports are defined at top level.

• Lazy-loaded LiteLLM usage consistency

  • Replaces direct litellm.acompletion(...) usage with _get_litellm().acompletion(...) in stream_completion, matching the existing lazy import/cache mechanism used elsewhere in the file.

• Net effect

  • Removes the CI lint failure source without changing AI service behavior or API surface.

# before
response = await litellm.acompletion(**completion_params)

# after
response = await _get_litellm().acompletion(**completion_params)

[Copilot]

Pull request overview

Fixes a CI lint failure in the AI service by restoring top-level import ordering and ensuring LiteLLM calls consistently go through the module’s lazy-loader path.

Changes:

• Moves get_settings into the module’s top-level import block to resolve E402 (import not at top of file).
• Replaces direct litellm.acompletion(...) usage in complete_stream with _get_litellm().acompletion(...) to match the existing lazy-load/cache pattern.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.