[NFC] Mark some DAE-internal passes as requiring less work (WebAssembly#7858)

The non-nullable overhead usually doesn't matter, but DAE can end up running
many internal pass iterations, many of whom are very lightweight otherwise,
so this adds up.

On a dart testcase I see, this makes DAE (the slowest pass there) 7% faster.

Author: kripken

src/passes/DeadArgumentElimination.cpp

@@ -94,6 +94,7 @@ using DAEFunctionInfoMap = std::unordered_map<Name, DAEFunctionInfo>;
 struct DAEScanner
   : public WalkerPass<PostWalker<DAEScanner, Visitor<DAEScanner>>> {
   bool isFunctionParallel() override { return true; }
+  bool modifiesBinaryenIR() override { return false; }
 
   std::unique_ptr<Pass> create() override {
     return std::make_unique<DAEScanner>(infoMap);

src/passes/param-utils.cpp

@@ -290,6 +290,9 @@ void localizeCallsTo(const std::unordered_set<Name>& callTargets,
                      std::function<void(Function*)> onChange) {
   struct LocalizerPass : public WalkerPass<PostWalker<LocalizerPass>> {
     bool isFunctionParallel() override { return true; }
+    // May add non-nullable locals, but fixups are never needed as they are
+    // immediately used in the code right after.
+    bool requiresNonNullableLocalFixups() override { return false; }
 
     std::unique_ptr<Pass> create() override {
       return std::make_unique<LocalizerPass>(callTargets, onChange);
@@ -335,6 +338,8 @@ void localizeCallsTo(const std::unordered_set<HeapType>& callTargets,
                      PassRunner* runner) {
   struct LocalizerPass : public WalkerPass<PostWalker<LocalizerPass>> {
     bool isFunctionParallel() override { return true; }
+    // See above.
+    bool requiresNonNullableLocalFixups() override { return false; }
 
     std::unique_ptr<Pass> create() override {
       return std::make_unique<LocalizerPass>(callTargets);