fix: store actual embedding model name in nodes metadata

Bug: embedding_model field in nodes table was showing hardcoded fallback
'jina-embeddings-v4' instead of the actual configured model (e.g.,
'qwen3-embedding:0.6b' for Ollama).

Root cause: annotate_node() function (line 2325) was not adding
embedding_model to node metadata attributes, so when nodes were converted
to SurrealNodeRecord, the embedding_model field remained None and fell
back to schema default ('jina-embeddings-v4').

Fix: Added embedding_model from self.embedding_model (which is correctly
set from config at line 474-478) to node metadata attributes at line 2347.

This ensures nodes table accurately reflects which embedding provider
was actually used during indexing.

Author: jruokola

crates/codegraph-mcp/src/indexer.rs

@@ -2341,6 +2341,10 @@ impl ProjectIndexer {
                 .attributes
                 .insert("domain".to_string(), domain.clone());
         }
+        // Add embedding model name for tracking which embedding provider was used
+        node.metadata
+            .attributes
+            .insert("embedding_model".to_string(), self.embedding_model.clone());
     }
 
     async fn persist_nodes_batch(&self, chunk: &[CodeNode]) -> Result<()> {