Simplify boilerplate for learning: stub inference and Postgres write

Removed ort/ndarray/image (ONNX Runtime inference) and sqlx (Postgres audit
write). inference::detect() and the Postgres half of reporter::report() are
now stubs with a comment for what real code goes there. Redis Streams
XREADGROUP loop and the Redis pub/sub alert stay real.

Verified with a clean cargo check --all-targets (0 warnings).

src/inference/mod.rs

@@ -1,4 +1,19 @@
-mod detector;
-mod preprocess;
+use serde::Serialize;
 
-pub use detector::{Detection, FailureDetector};
+#[derive(Debug, Clone, Serialize)]
+pub struct Detection {
+    pub label: String,
+    pub confidence: f32,
+}
+
+/// Stand-in for real ONNX Runtime inference (the `ort` crate, using a model
+/// trained on failure images). Always reports one fixed "failure" so the
+/// rest of the pipeline — reporting, acking the stream entry — has
+/// something to do and can be tested without a real model file.
+///
+/// Swap this out once you're comfortable with the pipeline around it:
+/// decode `frame_bytes` with the `image` crate, run it through an `ort`
+/// session, and turn the model's output into real `Detection`s.
+pub fn detect(_frame_bytes: &[u8]) -> Vec<Detection> {
+    vec![Detection { label: "spaghetti".to_string(), confidence: 0.91 }]
+}