Get LFQ item before capturing callstack.

This is to ensure that thread local structures have been properly
initialized (lock-free queue buffers are thread local), as capturing
callstack involves allocating memory from rpmalloc, which must be
initialized in each thread before allocation.

client/TracyProfiler.cpp

@@ -2825,10 +2825,9 @@ void Profiler::ReportTopology()
 void Profiler::SendCallstack( int depth, const char* skipBefore )
 {
 #ifdef TRACY_HAS_CALLSTACK
+    TracyLfqPrepare( QueueType::Callstack );
     auto ptr = Callstack( depth );
     CutCallstack( ptr, skipBefore );
-
-    TracyLfqPrepare( QueueType::Callstack );
     MemWrite( &item->callstackFat.ptr, (uint64_t)ptr );
     TracyLfqCommit;
 #endif