A whole bunch of optimization

• A lot of schedule optimization (specifically on miranda).
• Fix GPU backend emitting thread IDs in thread fork-joins.
• Add some sub edits.

hercules_cg/src/gpu.rs

@@ -1225,11 +1225,13 @@ namespace cg = cooperative_groups;
                             // because Fork basic block's init section already does gating
                             write!(
                                 w,
-                                "{}{} = (threadIdx.x % {}) / {};\n",
+                                "{}{} = (((threadIdx.x % {}) / {}) / ({})) % ({});\n",
                                 tabs,
                                 define_variable,
                                 use_thread_quota.unwrap(),
-                                use_thread_quota.unwrap() / parallel_factor.unwrap()
+                                use_thread_quota.unwrap() / parallel_factor.unwrap(),
+                                divide,
+                                modulo,
                             )?;
                         }
                     }