Use uv venv for embedding, pre-download model in Docker build

- Use /app/venv with uv instead of system python/pip
- Pre-download all-MiniLM-L6-v2 model during Docker build

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/kb.rs

@@ -101,7 +101,7 @@ impl KbManager {
 async fn compute_embeddings(texts: &[String]) -> Result<Vec<Vec<f32>>> {
     let input = serde_json::json!({ "texts": texts });
 
-    let mut child = tokio::process::Command::new("python3")
+    let mut child = tokio::process::Command::new("/app/venv/bin/python")
         .arg("/app/scripts/embed.py")
         .stdin(Stdio::piped())
         .stdout(Stdio::piped())