Don't output stdout from lld check in init-compiler.sh

[akoeplinger]

Another fallout from #8189 / #8219.

This caused an issue in dotnet/runtime#61668 because https://github.com/dotnet/runtime/blob/c00b06826ca4d333ef69d51e523ef7bd309b8631/src/libraries/System.Runtime.InteropServices/tests/TestAssets/NativeExports/NativeExports.csproj#L33 sources the init-compiler.sh and uses the stdout from it and we were outputting the stdout of the lld check which broke this.

Before:

# bash -c 'source "eng/common/native/init-compiler.sh" "eng/common/native/" x64 clang && echo $CC'
LLD 10.0.1 (compatible with GNU linkers)
/usr/bin/clang-10

After:

# bash -c 'source "eng/common/native/init-compiler.sh" "eng/common/native/" x64 clang && echo $CC'
/usr/bin/clang-10

[omajid]

FYI @janvorli @am11

[omajid]

Thanks!

[janvorli]

It seems that we should then explicitly mention in this scripts comments that the contract is that this script doesn't write anything to stdout. It is quite unexpected to me.

[akoeplinger]

@janvorli added a note

[am11]

Typical workflow is to modify files in runtime/eng/common, test it there and then issue PR in arcade (and optionally runtime if it's urgent and can't wait for arcade update). @akoeplinger, feel free to push that change to the runtime PR to unblock it, as the next sync will copy these files without a conflict.

@janvorli, agree that it's unexpected, e.g. in src/native/libs/build.sh, we treat pipeline failures as errors but not in coreclr/build-runtime.sh. We should use the same semantics repo-wide, so if it works for one component, it works everywhere alike. These disparities are not useful.

[akoeplinger]

@am11 we'll need another arcade bump anyway to fix another regression in Microsoft.DotNet.XunitExtensions (I'm preparing a PR right now).