[AMD] Handling denorms in lowering math.sqrt and math.sqrt_rn

[knwng]

In this commit, we handled the denorm flushing behaviors of math.sqrt and math.sqrt_rn. They read HIP_FTZ to determine whether denorms should be preserved or flushed to zero.

Backend	sqrt non-ftz	sqrt ftz	sqrt_rn non-ftz	sqrt_rn ftz
CUDA	sqrt.approx.f32	sqrt.approx.ftz.f32	sqrt.rn.f32	sqrt.rn.ftz.f32
AMD	scaling + llvm.amdgcn.sqrt.f32	llvm.amdgcn.sqrt.f32	llvm.sqrt.f32	llvm.amdgcn.rsq.f32 + mul + refinement

math.sqrt provides approximation of SQRT. In AMD backend, we use llvm.amdgcn.sqrt.f32, which provides direct access to v_sqrt_f32 and has 1ULP accuracy.

math.sqrt_rn provides IEEE-compliant result(round-to-nearest-or-even) of SQRT. Following the implementation in LLVM, we use extra refinement to get correctly rounded result.

New contributor declaration

• I am not making a trivial change, such as fixing a typo in a comment.

• I have written a PR description following these
  rules.

• I have run pre-commit run --from-ref origin/main --to-ref HEAD.

• Select one of the following.

  • I have added tests.
    • ✅ /test for lit tests
    • /unittest for C++ tests
    • /python/test for end-to-end tests
  • This PR does not need a test because FILL THIS IN.

• Select one of the following.

  • I have not added any lit tests.
  • The lit tests I have added follow these best practices,
    including the "tests should be minimal" section. (Usually running Python code
    and using the instructions it generates is not minimal.)

[zhanglx13]

LGTM. good to go after the comment is fixed