MemoryAwareTensorViewStrategy
Memory-pressure-aware tensor view strategy.
This strategy extends the default strategy with basic memory pressure detection. When memory pressure is high, it more aggressively prefers copying over views to allow garbage collection of parent tensors.