Optimize 2D+ loads as 1D when all columns are used

[dkoeplin]

e.g

val x = DRAM[Int](32, 32)

sram load x(0::32,0::32)

Should be treated as burst load of 1024 elements rather than 32 bursts of 32