docs(rust/face-recognition): document benchmarking.rs purpose and improve module comment

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

Author: marc0olo

rust/face-recognition/README.md

@@ -4,7 +4,7 @@ This example demonstrates running face detection and face recognition inside an
 
 The example consists of two canisters:
 
-- **backend** — embeds the Tract ONNX inference engine. Exposes endpoints for uploading ONNX model files in chunks, loading them into memory, detecting faces, computing face embeddings, and recognizing people.
+- **backend** — embeds the Tract ONNX inference engine. Exposes endpoints for uploading ONNX model files in chunks, loading them into memory, detecting faces, computing face embeddings, and recognizing people. Also exposes `run_detection` and `run_recognition` which run the models against a built-in test image and print the IC instruction count — useful for smoke-testing and capacity planning.
 - **frontend** — serves the web UI (HTML/JS/CSS).
 
 ## Models

rust/face-recognition/backend/src/benchmarking.rs

@@ -1,10 +1,17 @@
-// The code below is used for testing and benchmarking.
+// Benchmarking and smoke-test endpoints.
+//
+// run_detection() and run_recognition() run the respective models against a
+// hardcoded test image embedded in the Wasm binary. They print the IC instruction
+// count so you can measure the computational cost of each inference call.
+//
+// run_detection is a query (fast, single-node execution).
+// run_recognition is an update (replicated execution, consistent with recognize()).
 
 use crate::{onnx, Detection, Error, Recognition};
 
 const IMAGE: &'static [u8] = include_bytes!("../assets/image.png");
 
-/// Formats thousands for the specified `u64` integer (helper function).
+/// Formats a u64 with thousands separators, e.g. 1_234_567.
 fn fmt(n: u64) -> String {
     n.to_string()
         .as_bytes()