Bump spike buffer on GPU by 10x. Issue now and often requested.

arbor/backends/gpu/threshold_watcher.hpp

@@ -72,7 +72,7 @@ class threshold_watcher {
         v_prev_(num_cv),
         // TODO: allocates enough space for 10 spikes per watch.
         // A more robust approach might be needed to avoid overflows.
-        stack_(10*size(), context.gpu)
+        stack_(100*size(), context.gpu)
     {
         crossings_.reserve(stack_.capacity());
         // reset() needs to be called before this is ready for use