Optimizations for precision conversion operations in nGraph reference implementations (#3974)

* FP16->FP32 conversion vectorization

* int8->FP16 conversion vectorization in nGraph

.gitmodules

@@ -13,4 +13,8 @@
 [submodule "inference-engine/samples/thirdparty/gflags"]
 	path = inference-engine/samples/thirdparty/gflags
 	url = https://github.com/gflags/gflags.git
-	ignore = dirty
+	ignore = dirty
+[submodule "thirdparty/xbyak"]
+	path = thirdparty/xbyak
+	url = https://github.com/herumi/xbyak.git
+	ignore = dirty