[CRT_APITEST] Fix strlen test on x86

The code called winetest_setlocation, which calls strrchr, which clears the direction flag, so the test was succeeding on x86, even though strlen never changes it. On Windows it somehow depends on whether you compile the test on the command line or inside VS.
This change fixes the test by explicitly reading eflags before using winetest_ok and placing a MemoryBarrier in between.

Author: tkreuzer

modules/rostests/apitests/crt/strlen.c

@@ -32,7 +32,7 @@ Test_strlen(PFN_STRLEN pstrlen)
 {
     size_t len;
 #if defined(_M_IX86) || defined(_M_AMD64)
-    volatile uintptr_t eflags;
+    volatile uintptr_t eflags, eflags2;
     char *teststr = "a\0bcdefghijk";
 #endif
 
@@ -48,12 +48,9 @@ Test_strlen(PFN_STRLEN pstrlen)
     eflags = __readeflags();
     __writeeflags(eflags | EFLAGS_DF);
     len = pstrlen(teststr + 4);
-
-#ifdef _M_AMD64
-    ok((__readeflags() & EFLAGS_DF) != 0, "Direction flag in ELFAGS was changed.\n");
-#else
-    ok((__readeflags() & EFLAGS_DF) == 0, "Direction flag in ELFAGS was not changed.\n");
-#endif
+    eflags2 = __readeflags();
+    MemoryBarrier();
+    ok((eflags2 & EFLAGS_DF) != 0, "Direction flag in ELFAGS was changed: 0x%Ix -> 0x%Ix.\n", eflags | EFLAGS_DF, eflags2);
     __writeeflags(eflags);
 
     /* Only test this for the exported versions, intrinsics might do it