[SYCL] Fix hang in deallocateStreams function (#2796)

This patch fixes hang caused by data race in `scheduler.cpp`'s
`deallocateStreams` function: data of `StreamsToDeallocate` vector is
corrupted by another thread, and due to this, it was a hang in
`Scheduler::deallocateStreamBuffers` function when
`StreamBuffersPool[Impl]` had been being deleted.

Author: dm-vodopyanov

sycl/source/detail/scheduler/scheduler.cpp

@@ -276,13 +276,13 @@ void Scheduler::enqueueLeavesOfReqUnlocked(const Requirement *const Req) {
 void Scheduler::allocateStreamBuffers(stream_impl *Impl,
                                       size_t StreamBufferSize,
                                       size_t FlushBufferSize) {
-  std::lock_guard<std::mutex> lock(StreamBuffersPoolMutex);
+  std::lock_guard<std::recursive_mutex> lock(StreamBuffersPoolMutex);
   StreamBuffersPool.insert(
       {Impl, new StreamBuffers(StreamBufferSize, FlushBufferSize)});
 }
 
 void Scheduler::deallocateStreamBuffers(stream_impl *Impl) {
-  std::lock_guard<std::mutex> lock(StreamBuffersPoolMutex);
+  std::lock_guard<std::recursive_mutex> lock(StreamBuffersPoolMutex);
   delete StreamBuffersPool[Impl];
   StreamBuffersPool.erase(Impl);
 }
@@ -302,7 +302,7 @@ Scheduler::~Scheduler() {
   // the kernel. Otherwise resources for stream will not be released, issue a
   // warning in this case.
   if (pi::trace(pi::TraceLevel::PI_TRACE_BASIC)) {
-    std::lock_guard<std::mutex> lock(StreamBuffersPoolMutex);
+    std::lock_guard<std::recursive_mutex> lock(StreamBuffersPoolMutex);
     if (!StreamBuffersPool.empty())
       fprintf(
           stderr,

sycl/source/detail/scheduler/scheduler.hpp

@@ -766,7 +766,7 @@ class Scheduler {
   friend class stream_impl;
 
   // Protects stream buffers pool
-  std::mutex StreamBuffersPoolMutex;
+  std::recursive_mutex StreamBuffersPoolMutex;
 
   // We need to store a pointer to the structure with stream buffers because we
   // want to avoid a situation when buffers are destructed during destruction of