perf: eliminate per-request SQLite queries and optimize proxy latency

- Add in-memory ModelConfigCache (30s refresh, explicit invalidation)
  replacing 2 SQLite queries per request (model lookup + cost override)
- Configure all 5 provider HTTP clients with proper timeouts (300s),
  connection pooling (4 idle/host, 90s idle timeout), and TCP keepalive
- Move client_usage update to tokio::spawn in non-streaming path
- Use fast chars/4 heuristic for token estimation on large inputs (>1KB)
- Generate single UUID/timestamp per SSE stream instead of per chunk
- Add shared LazyLock<Client> for image fetching in multimodal module
- Add proxy overhead timing instrumentation for both request paths
- Fix test helper to include new model_config_cache field

src/providers/ollama.rs

@@ -14,8 +14,16 @@ pub struct OllamaProvider {
 
 impl OllamaProvider {
     pub fn new(config: &crate::config::OllamaConfig, app_config: &AppConfig) -> Result<Self> {
+        let client = reqwest::Client::builder()
+            .connect_timeout(std::time::Duration::from_secs(5))
+            .timeout(std::time::Duration::from_secs(300))
+            .pool_idle_timeout(std::time::Duration::from_secs(90))
+            .pool_max_idle_per_host(4)
+            .tcp_keepalive(std::time::Duration::from_secs(30))
+            .build()?;
+
         Ok(Self {
-            client: reqwest::Client::new(),
+            client,
             config: config.clone(),
             pricing: app_config.pricing.ollama.clone(),
         })