[red-knot] Module resolver has incorrect behaviour if a single-file module and potential namespace package coexist

[AlexWaygood]

The following test should pass, but fails:

diff --git a/crates/red_knot_module_resolver/src/resolver.rs b/crates/red_knot_module_resolver/src/resolver.rs
index 71b807871..5decee0c6 100644
--- a/crates/red_knot_module_resolver/src/resolver.rs
+++ b/crates/red_knot_module_resolver/src/resolver.rs
@@ -1140,4 +1140,21 @@ mod tests {
             system_path_to_file(&db, stdlib.join("functools.pyi"))
         );
     }
+
+    #[test]
+    fn single_file_takes_priority_over_namespace_package() {
+        const SRC: &[FileSpec] = &[("foo.py", "x = 1"), ("foo/bar.py", "x = 2")];
+
+        let TestCase { db, src, .. } = TestCaseBuilder::new().with_src_files(SRC).build();
+
+        let foo_module_name = ModuleName::new_static("foo").unwrap();
+        let foo_bar_module_name = ModuleName::new_static("foo.bar").unwrap();
+
+        // `foo.py` takes priority over the `foo` directory;
+        // `foo.bar` isn't recognised as a module:
+        let foo_module = resolve_module(&db, foo_module_name.clone()).unwrap();
+        let foo_path = src.join("foo.py");
+        assert_eq!(&foo_path, foo_module.file().path(&db));
+        assert_eq!(resolve_module(&db, foo_bar_module_name.clone()), None);
+    }
 }

At runtime, Python considers that a foo module exists and considers that a foo.bar module does not exist (it does not view the foo module as a package, as the foo.py file takes priority over the foo directory). The red-knot module resolver currently has the opposite understanding: it believes that there is no module foo, and that there is a module foo.bar.