qemu: signal correct exit code to QEMU

There were a few issues that were causing qemu-system-arm and
qemu-system-riscv to give the wrong exit codes. They are in fact capable
of exiting with 0 or 1 signalled from the running application, but this
functionality wasn't used. This commit changes this in the following
ways:

  * It fixes SemiHosting codes, which were incorrectly written in
    decimal while they should have been written in hexadecimal (oops!).
  * It modifies all the baremetal main functions (aka reset handlers) to
    exit with `exit(0)` instead of `abort()`.
  * It changes `syscall.Exit` to call `exit(code)` instead of `abort()`
    on baremetal targets.
  * It adds these new exit functions where necessary, implemented in a
    way that signals the correct exit status if running under QEMU.

All in all, this means that `tinygo test` doesn't have to look at the
output of a test to determine the outcome. It can simply look at the
exit code.

src/runtime/runtime_cortexm_qemu.go

@@ -23,8 +23,7 @@ func main() {
 	run()
 
 	// Signal successful exit.
-	arm.SemihostingCall(arm.SemihostingReportException, arm.SemihostingApplicationExit)
-	abort()
+	exit(0)
 }
 
 func ticksToNanoseconds(ticks timeUnit) int64 {
@@ -56,8 +55,16 @@ func waitForEvents() {
 }
 
 func abort() {
-	// Signal an abnormal exit.
-	arm.SemihostingCall(arm.SemihostingReportException, arm.SemihostingRunTimeErrorUnknown)
+	exit(1)
+}
+
+func exit(code int) {
+	// Exit QEMU.
+	if code == 0 {
+		arm.SemihostingCall(arm.SemihostingReportException, arm.SemihostingApplicationExit)
+	} else {
+		arm.SemihostingCall(arm.SemihostingReportException, arm.SemihostingRunTimeErrorUnknown)
+	}
 
 	// Lock up forever (should be unreachable).
 	for {