Send ONNX Runtime's session log to the app's log

A native session's messages went to ONNX Runtime's stdio logger, which is
nowhere once the app is launched from a menu — and what a provider says
while partitioning a graph (nodes taken, operators declined, a library that
failed to load) is most of what a failed rung tells you. Each session now
forwards them to `log` under the target `onnxruntime`: warnings always,
the runtime's info lines at `debug`, its verbose lines at `trace`.

core/dr-inference-engine/src/session.rs

@@ -19,6 +19,9 @@ pub fn build(rung: Rung, role: Role, bytes: &[u8], cfg: &Config) -> ort::Result<
     // `stack_tensors`) — a panic across the C API, which is an abort. The
     // app never asked tract for that and does not start now.
     let mut b = Session::builder()?.with_intra_threads(threads(cfg))?;
+    if crate::api::runtime().is_native() {
+        b = with_runtime_log(b)?;
+    }
     // A Hexagon session loads the compiled context when there is one and
     // compiles it from the model when there is not; the engine thread is
     // what makes the second case rare (§6).
@@ -37,6 +40,39 @@ pub fn build(rung: Rung, role: Role, bytes: &[u8], cfg: &Config) -> ort::Result<
     }
 }
 
+/// Send the runtime's own messages for this session to `log`, under the
+/// target `onnxruntime`, instead of to ONNX Runtime's stdio logger.
+///
+/// Its stderr is nowhere once the app is launched from a menu, and what a
+/// provider says while it partitions a graph — how many nodes it took, which
+/// operator it declined, the library it failed to load — is most of what a
+/// failed rung tells you (docs/dev/inference.md §4). The level follows the
+/// filter: warnings always, `debug` adds the runtime's info lines (the
+/// partition counts), `trace` its verbose ones (every node placement).
+fn with_runtime_log(
+    b: ort::session::builder::SessionBuilder,
+) -> ort::Result<ort::session::builder::SessionBuilder> {
+    use ort::logging::LogLevel;
+    let level = if log::log_enabled!(target: "onnxruntime", log::Level::Trace) {
+        LogLevel::Verbose
+    } else if log::log_enabled!(target: "onnxruntime", log::Level::Debug) {
+        LogLevel::Info
+    } else {
+        LogLevel::Warning
+    };
+    let forward = |level: LogLevel, _category: &str, _id: &str, location: &str, message: &str| {
+        let level = match level {
+            LogLevel::Verbose => log::Level::Trace,
+            LogLevel::Info => log::Level::Debug,
+            LogLevel::Warning => log::Level::Warn,
+            LogLevel::Error | LogLevel::Fatal => log::Level::Error,
+        };
+        log::log!(target: "onnxruntime", level, "{message} ({location})");
+    };
+    Ok(b.with_logger(std::sync::Arc::new(forward))?
+        .with_log_level(level)?)
+}
+
 /// The intra-op pool: what the config says, else the cores less two for
 /// the compositor and the decoder (§9). tract ignores it.
 fn threads(cfg: &Config) -> usize {