Move memory barrier in UnlockBufHdr to before releasing the lock.

This bug appears to have been introduced late in the development of
48354581a4 ("Allow Pin/UnpinBuffer to operate in a lockfree
manner.").

Found while debugging a bug which turned out to be independent of the
commit mentioned above.

Backpatch: -

src/include/storage/buf_internals.h

@@ -242,8 +242,8 @@ extern PGDLLIMPORT LWLockMinimallyPadded *BufferIOLWLockArray;
 extern uint32 LockBufHdr(BufferDesc *desc);
 #define UnlockBufHdr(desc, s)	\
 	do {	\
-		pg_atomic_write_u32(&(desc)->state, (s) & (~BM_LOCKED)); \
 		pg_write_barrier(); \
+		pg_atomic_write_u32(&(desc)->state, (s) & (~BM_LOCKED)); \
 	} while (0)