[llvm] [AMDGPU] Allow shinking instruction with dead sdst (PR #68028)

[Stanislav Mekhanoshin via llvm-commits]

rampitec wrote:

> For GFX, I quickly tried this on Navi21 shaders and saw no shift in _e32 versus _e64. But I might have missed something.

For Navi21, note this code in the shrink pass:

```
      // fold an immediate into the shrunk instruction as a literal operand. In
      // GFX10 VOP3 instructions can take a literal operand anyway, so there is
      // no advantage to doing this.
      if (ST->hasVOP3Literal() &&
          !MF.getProperties().hasProperty(
              MachineFunctionProperties::Property::NoVRegs))
        continue;
```

Which means, it is disabled for Navi.

https://github.com/llvm/llvm-project/pull/68028