refactor(indexer): consolidate embedding initialization to use config

Architectural fix: Symbol resolution now uses the same configuration path
as main indexing instead of creating a separate EmbeddingGenerator with
environment variables.

Before: ai_resolve_symbol() called with_auto_from_env() - different init path
After: ai_resolve_symbol() calls with_config(&self.global_config) - same as main indexing

This ensures:
1. Single source of truth for embedding configuration
2. Consistent provider selection across all code paths
3. No duplicate initialization logic
4. Easier maintenance and testing

The indexer already has access to global_config in the ProjectIndexer struct,
so this is a simple one-line change that eliminates architectural inconsistency.

Author: jruokola

crates/codegraph-mcp/src/indexer.rs

@@ -1674,8 +1674,8 @@ impl ProjectIndexer {
         use codegraph_vector::{search::SemanticSearch, EmbeddingGenerator};
         use std::sync::Arc;
 
-        // Create a simple embedding for the target symbol
-        let embedder = EmbeddingGenerator::with_auto_from_env().await;
+        // Use same config as main indexing for consistency
+        let embedder = EmbeddingGenerator::with_config(&self.global_config).await;
         if let Ok(target_embedding) = embedder.generate_text_embedding(target_symbol).await {
             // Find the most similar symbol in our symbol map using cosine similarity
             let mut best_match: Option<(NodeId, f32)> = None;