[DNM] Add hardware loop memcpy optimization

Created by: flip1995

DNM: This PR is awaiting testing on the hardware before it can get merged.

TODO:

• There are now more contraints on HWLPs than just "no compressed instructions". Those have to be addressed before merging/testing this.
• Fix bug, when compiled with -Os:

  typedef unsigned char uint8_t;
  
  struct S {
      uint8_t b[N];
  };
  
  extern struct S a, b;
  
  uint8_t f (int i)
  {
      if(i)
          a = b;
  
      return a.b[0];
  }

  When compiled with -Os this tests moves the dummy basic block containing the nop instruction.

r? @simonpcook