Misc. improvements for GPU testing

• Draw functions w/ colors per-device in graphviz visualization.
• Fix bugs in GCM and loop nesting to allow forks to get to codegen.
• Add GPU schedules for dot and matmul tests that contain forks.
  • Currently commented out in the build scripts, just switch which schedule_in_src line is commented out in both to actually use the GPU schedule when using --features=cuda.

hercules_ir/src/dot.rs

@@ -219,7 +219,13 @@ fn write_subgraph_header<W: Write>(
     } else {
         write!(w, "label=\"{}\"\n", function.name)?;
     }
-    write!(w, "bgcolor=ivory4\n")?;
+    let color = match function.device {
+        Some(Device::LLVM) => "paleturquoise1",
+        Some(Device::CUDA) => "darkseagreen1",
+        Some(Device::AsyncRust) => "peachpuff1",
+        None => "ivory2",
+    };
+    write!(w, "bgcolor={}\n", color)?;
     write!(w, "cluster=true\n")?;
     Ok(())
 }