dma: dmatest: fix potential sign bug

'cnt' is unsigned, so this code may become wrong in future as
dmatest_add_threads() can return error code:

	cnt = dmatest_add_threads(dtc, DMA_MEMCPY);
	thread_count += cnt > 0 ? cnt : 0;
		        ^^^^^^^

Now it can return only -EINVAL if and only if second argument of
dmatest_add_threads() is not one of DMA_MEMCPY, DMA_XOR, DMA_PQ.
So, now it is not wrong but may become wrong in future.

The semantic patch that finds this problem (many false-positive results):
(http://coccinelle.lip6.fr/)

// <smpl>
@ r1 @
identifier f;
@@
int f(...) { ... }

@@
identifier r1.f;
type T;
unsigned T x;
@@

*x = f(...)
 ...
*x > 0

Signed-off-by: Kulikov Vasiliy <segooon@gmail.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

drivers/dma/dmatest.c

@@ -540,7 +540,7 @@ static int dmatest_add_channel(struct dma_chan *chan)
 	struct dmatest_chan	*dtc;
 	struct dma_device	*dma_dev = chan->device;
 	unsigned int		thread_count = 0;
-	unsigned int		cnt;
+	int cnt;
 
 	dtc = kmalloc(sizeof(struct dmatest_chan), GFP_KERNEL);
 	if (!dtc) {