Fix enabling optimizations for LLVM 16

The changes to how passes are run in the new pass manager means simply
setting the optimization level isn't enough, instead you have to add the
pass group "default<N>" where "N" is "O1", "O2" or "O3". This ensures
that for example the module inliner is correctly enabled when using
--opt=aggressive.

The resulting performance still isn't great as LLVM doesn't
inline/optimize across modules, but it's better than --opt=aggressive
effectively doing very little.

compiler/src/llvm/passes.rs

@@ -617,7 +617,11 @@ impl<'a> Worker<'a> {
     fn run_passes(&self, module: &Module) {
         let layout = self.machine.get_target_data().get_data_layout();
         let opts = PassBuilderOptions::create();
-        let passes = &["mem2reg"];
+        let passes = if let Opt::Aggressive = self.shared.state.config.opt {
+            &["default<O3>"]
+        } else {
+            &["mem2reg"]
+        };
 
         module.set_data_layout(&layout);
         module.set_triple(&self.machine.get_triple());