feat(chat): implement speculative web search and embeddings commands

- Added speculative search functionality to enhance query processing by detecting when a web search may be needed based on user input patterns.
- Integrated commands for managing embeddings, including downloading models, checking availability, calculating relevance scores, generating embeddings, and pruning context.
- Optimized chat streaming by buffering tokens to reduce overhead and improve responsiveness.
- Implemented lazy initialization for the headless browser to conserve memory until needed.
- Enhanced web scraping capabilities with a static scraping method for faster content retrieval.

src-tauri/Cargo.toml

@@ -47,3 +47,10 @@ tauri-plugin-autostart = "2"
 dirs = "5.0"
 rfd = "0.14"
 rusqlite = { version = "0.31", features = ["bundled"] }
+
+# ONNX Runtime para embeddings locais (ranking de relevância)
+# Usando load-dynamic para evitar conflito de RuntimeLibrary (MD vs MT) no Windows
+# entre ort (dynamic) e tokenizers/esaxx-rs (static)
+ort = { version = "2.0.0-rc.10", features = ["load-dynamic"] }
+ndarray = "0.16"
+tokenizers = "0.21"