Add tag cloud feature with k-means clustering and CUDA support

Introduces an Explore view with a tag cloud that clusters image embeddings
using cosine k-means and labels clusters via vocabulary-nearest-neighbour
CLIP matching. Vocabulary embeddings are disk-cached (FNV hash-keyed) to
avoid redundant inference. Enables CUDA for candle dependencies and adds a
build.rs check that surfaces a clear error when the toolkit is missing.

src-tauri/build.rs

@@ -1,3 +1,28 @@
 fn main() {
-    tauri_build::build()
+    tauri_build::build();
+
+    // When the candle-cuda feature is enabled, verify the CUDA Toolkit is present
+    // so the developer gets a clear message rather than an obscure cudarc/cc error.
+    if std::env::var("CARGO_FEATURE_CANDLE_CUDA").is_ok() {
+        let cuda_path = std::env::var("CUDA_PATH"); // Set by the CUDA Toolkit installer on Windows
+        let nvcc_found = std::process::Command::new("nvcc")
+            .arg("--version")
+            .output()
+            .map(|o| o.status.success())
+            .unwrap_or(false);
+
+        if cuda_path.is_ok() || nvcc_found {
+            if let Ok(path) = &cuda_path {
+                println!("cargo:warning=CUDA Toolkit found at {path} — GPU acceleration enabled.");
+            } else {
+                println!("cargo:warning=CUDA Toolkit (nvcc) found in PATH — GPU acceleration enabled.");
+            }
+        } else {
+            println!("cargo:warning=━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━");
+            println!("cargo:warning=  candle-cuda feature is enabled but no CUDA Toolkit found.");
+            println!("cargo:warning=  Install CUDA Toolkit from https://developer.nvidia.com/cuda-downloads");
+            println!("cargo:warning=  Or build without GPU support: cargo build --no-default-features");
+            println!("cargo:warning=━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━");
+        }
+    }
 }