RFR: Add read barriers to C1's vectorizedMismatch() intrinsic

[Roman Kennke]

The testcase for the JNI critical stuff turned up a bug (yay testdriven
development!). C1's vectorizedMismatch() intrinsic was lacking read
barriers (the intrinsic was probably recently added).

This patch adds them, and includes a testcase (very similar to the JNI
critical testcase, but calls only C1 and System.arraycopy() instead of
JNI).

http://cr.openjdk.java.net/~rkennke/c1_vect_mismatch/webrev.00/

Ok to go?

Roman