Speed up cameras to 25fps by parallelizing solves

Running solves in parallel is much more efficient than parallelizing
each one. This lets us comfortably process images at 25fps.

Change-Id: I572ae60cf70b183d2a7716fe53c80475f862f24d
diff --git a/y2019/vision/target_finder.h b/y2019/vision/target_finder.h
index 3da5373..ff4c3f7 100644
--- a/y2019/vision/target_finder.h
+++ b/y2019/vision/target_finder.h
@@ -59,6 +59,8 @@
       const std::vector<TargetComponent> component_list, bool verbose);
 
   // Given a target solve for the transformation of the template target.
+  //
+  // This is safe to call concurrently.
   IntermediateResult ProcessTargetToResult(const Target &target, bool verbose);
 
   // Returns true if a target is good, and false otherwise.  Picks the 4 vs 8