fix: wire all LLM clients to api.riotpiao.com gateway

- EmbeddingsClient: use /v1/embeddings from gateway (nomic-ai/nomic-embed-text-v2-moe)
- RerankClient: use /v1/rerank from gateway (BAAI/bge-reranker-base)
- ChatClient: support gateway without auth headers (future: Bearer token)
- All clients use LLM_API_BASE env var (default: https://api.riotpiao.com)
- Ready for Kubernetes deployment with proper API routing

crates/mem-cli/src/http_server.rs

@@ -1,6 +1,6 @@
 use actix_web::{web, App, HttpServer, HttpResponse, HttpRequest, middleware::Logger};
 use anyhow::Result;
-use mem_llm::{ChatClient, EmbeddingsClient, RerankClient};
+use mem_llm::{EmbeddingsClient, RerankClient};
 use mem_store::{init_schema, VectorStore};
 use serde_json::json;
 use sqlx::PgPool;
@@ -49,9 +49,7 @@ pub async fn start_server(port: u16, api_key: String, database_url: &str) -> Res
     let vector_store = Arc::new(VectorStore::new(pool.clone()));
     let embeddings = Arc::new(EmbeddingsClient::from_env()?);
     let ingest_worker = Arc::new(IngestWorker::new(pool.clone(), (*embeddings).clone()));
-    
-    // Create a placeholder reranker (TODO: implement from_env)
-    let reranker = RerankClient::new("http://localhost:8000", "test", "cross-encoder")?;
+    let reranker = RerankClient::from_env()?;
     let query_worker = Arc::new(QueryWorker::new(VectorStore::new(pool.clone()), (*embeddings).clone(), reranker));
 
     let state = web::Data::new(AppState {