Improve error reporting for intrinsic methods with type mismatched arguments

[pow2clk]

In the course of writing tests for the fix to #2693 and #818, I discovered that intrinsic methods have their own error reporting path that fix doesn't resolve.

https://godbolt.org/z/ExYrj31z9

Texture2D<float4> tex;
SamplerComparisonState samp;

float4 main(float2 coord : C) : SV_Target {
  return tex.Sample(samp, coord);
}

The reporting attempts to report why both variants don't match, but the reporting for the arg type mismatch is elided due to incomplete error reporting from the HLSL code, so the only error message is for the second version, which takes more arguments, leading to this baffling error:

<source>:5:14: error: no matching member function for call to 'Sample'
  return tex.Sample(samp, coord);
         ~~~~^~~~~~
<source>:5:14: note: candidate function template not viable: requires 3 arguments, but 2 were provided
<source>:5:14: note: candidate function template not viable: requires 4 arguments, but 2 were provided
<source>:5:14: note: candidate function template not viable: requires 5 arguments, but 2 were provided

[llvm-beanz]

Gregory Roth (@pow2clk) the code provided here compiles successfully now (godbolt).

That said, I think the underlying issue still likely exists. Do you think this is likely to be something we will work on or should we close it out?

[pow2clk]

The issue persists even though the example doesn't produce it anymore. I can come up with a different example. I think we should keep awareness of it because we've had some efforts to make better diagnostics for specific cases that this would make unnecessary.

[pow2clk]

Exampled updated to a quite plausible mistake to make.