Reserve scratch on first queue allocation

Some workloads running on multi-GPU create 1 process per GPU. So each
process creates a GPU agent on every GPU, but will only create queues on
one GPU. This would cause un-necessary scratch reservation.

Change-Id: I50a216f0bcc0b5f707f3943147390b0ecec1ac22

runtime/hsa-runtime/core/inc/amd_gpu_agent.h

@@ -343,6 +343,8 @@ class GpuAgent : public GpuAgentInt {
   // @brief returns true if agent uses MES scheduler
   __forceinline const bool isMES() const { return (isa_->GetMajorVersion() >= 11) ? true : false; };
 
+  void ReserveScratch();
+
   void Trim() override;
 
   const std::function<void*(size_t size, size_t align, core::MemoryRegion::AllocateFlags flags)>&