perf: With StateMap the memory for states are cut in half

src/dfa.rs

@@ -1517,7 +1517,7 @@ impl<'a> Fsm<'a> {
         // so we can find it later.
         self.cache.size +=
             self.cache.trans.state_heap_size()
-            + (2 * state.data.len())
+            + state.data.len()
             + (2 * mem::size_of::<State>())
             + mem::size_of::<StatePtr>();
         self.cache.compiled.insert(state, si);