ScopeCheck sees DRAM dimension access as illegal input

[mattfel1]

If I try to do something like this, it sees the input.rows and input.cols accesses as illegal, since they are RegRead of ArgIns created by the DRAM. I put a temporary hack in ScopeCheck.scala but this should probably be fixed for real

  def Convolution(output: DRAM2[T], 
                      input: DRAM2[T],
                      ...): Unit = {

...
    Foreach(input.rows by rowstride){row =>
      Foreach(input.cols by coltile){col => 
...

[dkoeplin]

Fixed - just have to move Scope Check after we fix this in the IR