[DirectX] Implement Shader Flag Analysis for RequiresGroup

[bharadwajy]

DXC sets the flag m_bRequiresGroup to indicate a function accesses group shared memory. It is set if any of the function instruction's type or operand type being a pointer into thread group shared memory - per this and this.

1. Need to add SHADER_FEATURE_FLAG(32, 40, RequiresGroup, "Function accesses thread group shared memory") to llvm/include/llvm/BinaryFormat/DXContainerConstants.def.
2. Set RequiresGroup if any of the function instruction's type or operand type being a pointer into thread group shared memory.

Clarification needed: Does clang use a separate address space or some other mechanism to represent thread group shared memory?

AC:

Add tests to verify correctness of analysis.