[Feature]: `semantic_search_nodes` ignores `CRG_OPENAI_*` env vars and always falls back to local provider when `provider` is not explicitly passed

[destitutus]

Problem

Version

2.3.5

Problem

When CRG_OPENAI_API_KEY, CRG_OPENAI_BASE_URL, and CRG_OPENAI_MODEL are configured (e.g. for a local Ollama instance), calling semantic_search_nodes without an explicit provider argument always falls back to keyword-only search instead of using the configured OpenAI-compatible backend.

The root cause is in get_provider() (embeddings.py): when provider=None, the function immediately falls through to the # Default: local branch without checking whether OpenAI env vars are present:

# Default: local
if not _check_available():
    return None
return LocalEmbeddingProvider(model_name=model)

This means embeddings built with --provider openai are effectively unreachable unless every caller explicitly passes provider="openai" ? including all MCP tool invocations from AI assistants, which typically omit optional parameters.

Expected behavior

If CRG_OPENAI_API_KEY and CRG_OPENAI_BASE_URL are set, get_provider(provider=None) should default to the OpenAI-compatible provider rather than local.

Proposed solution

Suggested fix

def get_provider(
    provider: str | None = None,
    model: str | None = None,
) -> EmbeddingProvider | None:
    if provider is None and os.environ.get("CRG_OPENAI_API_KEY") and os.environ.get("CRG_OPENAI_BASE_URL"):
        provider = "openai"
    if provider == "openai":
        ...

This makes the provider selection consistent with how the OpenAI env vars are already used elsewhere in the codebase ? no new env vars needed.

Affected area

None

Alternatives considered

No response

Additional context

No response