PC Sampling: Add s_nop prior to s_sendmeg

Add s_nop before s_sendmsg. This is required because the HW does not
check for dependencies for SALU writes to M0.

Section 4.5: Manually Inserted Wait States (NOPs)
"AMD Instinct MI200" Instruction Set Architecture
https://www.amd.com/content/dam/amd/en/documents/instinct-tech-docs/instruction-set-architectures/instinct-mi200-cdna2-instruction-set-architecture.pdf

Change-Id: I90f503e3cc80cd29eab8bafa2565699461654055
Signed-off-by: Chris Freehill <cfreehil@amd.com>

runtime/hsa-runtime/core/runtime/trap_handler/trap_handler.s

@@ -393,6 +393,7 @@ trap_entry:
   s_waitcnt             lgkmcnt(0)
   s_mov_b32             ttmp13, m0                      // save off m0
   s_mov_b32             m0, ttmp7                       // put ID into message payload
+  s_nop                 0x0                             // Manually inserted wait states
   s_sendmsg             sendmsg(MSG_INTERRUPT)          // send interrupt message
   s_waitcnt             lgkmcnt(0)                      // wait for message to be sent
   s_mov_b32             m0, ttmp13                      // restore m0