Upgrade halide-llvm to 23.0.0.dev85505+g69780be1 (#9028)

* Upgrade halide-llvm to 23.0.0.dev85505+g69780be1

* Remove usage of NoNansFPMath after LLVM 22 (#9039)

This was removed upstream and was a no-op when it got removed, so we can
simply hide these options behind an if-def.

---------

Co-authored-by: halide-ci[bot] <266445882+halide-ci[bot]@users.noreply.github.com>
Co-authored-by: Aiden Grossman <aidengrossman@google.com>

Author: halide-ci[bot]

src/CodeGen_Internal.cpp

@@ -614,8 +614,8 @@ void get_target_options(const llvm::Module &module, llvm::TargetOptions &options
     options.AllowFPOpFusion = per_instruction_fast_math_flags ? llvm::FPOpFusion::Strict : llvm::FPOpFusion::Fast;
 #if LLVM_VERSION < 230
     options.NoInfsFPMath = !per_instruction_fast_math_flags;
-#endif
     options.NoNaNsFPMath = !per_instruction_fast_math_flags;
+#endif
     options.HonorSignDependentRoundingFPMathOption = !per_instruction_fast_math_flags;
     options.NoZerosInBSS = false;
     options.GuaranteedTailCallOpt = false;

src/CodeGen_PTX_Dev.cpp

@@ -630,8 +630,8 @@ vector<char> CodeGen_PTX_Dev::compile_to_src() {
     options.AllowFPOpFusion = CodeGen_GPU_Dev::any_strict_float ? llvm::FPOpFusion::Strict : llvm::FPOpFusion::Fast;
 #if LLVM_VERSION < 230
     options.NoInfsFPMath = !CodeGen_GPU_Dev::any_strict_float;
-#endif
     options.NoNaNsFPMath = !CodeGen_GPU_Dev::any_strict_float;
+#endif
     options.HonorSignDependentRoundingFPMathOption = !CodeGen_GPU_Dev::any_strict_float;
     options.NoZerosInBSS = false;
     options.GuaranteedTailCallOpt = false;