Don't automagically force lld > gold > bfd. Leave it up to the user.

In particular, avoids issues with LTO enabled (via the flag/scons option)
where using GCC as compiler, as lld can't do LTO with GCC.

https://bugs.gentoo.org/769986
--- a/SConstruct
+++ b/SConstruct
@@ -3109,20 +3109,6 @@ def doConfigure(myenv):
         myenv.Append( CCFLAGS=["/Zc:inline"])
 
     if myenv.ToolchainIs('gcc', 'clang'):
-        # This tells clang/gcc to use the gold linker if it is available - we prefer the gold linker
-        # because it is much faster. Don't use it if the user has already configured another linker
-        # selection manually.
-        if not any(flag.startswith('-fuse-ld=') for flag in env['LINKFLAGS']):
-
-            # lld has problems with separate debug info on some platforms. See:
-            # - https://bugzilla.mozilla.org/show_bug.cgi?id=1485556
-            # - https://bugzilla.mozilla.org/show_bug.cgi?id=1485556
-            if get_option('separate-debug') == 'off':
-                if not AddToLINKFLAGSIfSupported(myenv, '-fuse-ld=lld'):
-                    AddToLINKFLAGSIfSupported(myenv, '-fuse-ld=gold')
-            else:
-                AddToLINKFLAGSIfSupported(myenv, '-fuse-ld=gold')
-
         # Usually, --gdb-index is too expensive in big static binaries, but for dynamic
         # builds it works well.
         if link_model.startswith("dynamic"):