[Comgr] Pass -no-disable-free to Clang

The clang driver passes -disable-free to speedup clang's process by
skipping destructors at exit.

This is problematic for Comgr since we do not want to leak memory at
every compilation.

Before this patch, we worked around this issue by removing the -disable-free flag from the the Job
arguments.

In this patch, we pass -no-disable-free to the driver.

When we pass -no-disable-free, both flags are passed but
-no-disable-free is picked since it's the last option related to
disable-free in the command line. Then, we can remove the previous hack.

Author: jmmartinez

amd/comgr/src/comgr-compiler.cpp

@@ -649,12 +649,6 @@ amd_comgr_status_t executeCommand(const Command &Job, raw_ostream &LogS,
   Argv.append(Arguments.begin(), Arguments.end());
   Argv.push_back(nullptr);
 
-  // By default clang driver will ask CC1 to leak memory.
-  auto *IT = find(Argv, StringRef("-disable-free"));
-  if (IT != Argv.end()) {
-    Argv.erase(IT);
-  }
-
   clearLLVMOptions();
 
   if (Argv[1] == StringRef("-cc1")) {
@@ -900,6 +894,10 @@ amd_comgr_status_t AMDGPUCompiler::processFile(DataObject *Input,
     Argv.push_back(Flag);
   }
 
+  // By default clang driver will ask CC1 to leak memory.
+  Argv.push_back("-Xclang");
+  Argv.push_back("-no-disable-free");
+
   Argv.push_back(InputFilePath);
 
   Argv.push_back("-o");