remove SinkingPass from pass pipeline

SinkingPass is not used by Clang and seems to be only used
as part of the AMDGPU pipeline (after the CFGStructurizer).

I have seen examples where it creates substantial worse code,
especially for the NVPTX, where it clusters loads and moves arithmetic
operations far away, causing a much higher register usage.

src/jitlayers.cpp

@@ -213,7 +213,6 @@ void addOptimizationPasses(legacy::PassManagerBase *PM, int opt_level,
 
  // Run instcombine after redundancy elimination to exploit opportunities
  // opened up by them.
- PM->add(createSinkingPass()); ////////////// ****
 #if JL_LLVM_VERSION < 70000
  // This pass is a subset of InstructionCombiningPass in LLVM 7
  // and therefore not required.