More defence in the thread unwind path.

We were spilling the trusted stack pointer to the csp slot in the
thread's register save area.  This shouldn't matter because you can't
resume a thread that has exited, but it's good defence in depth to not
do this and spill 0 instead.

sdk/core/switcher/entry.S

@@ -777,8 +777,11 @@ exception_entry_asm:
 	// Value 24 is reserved for custom use.
 .Lset_mcause_and_exit_thread:
 	csrw               mcause, 24
-	// The thread exit code expects the trusted stack pointer to be in csp.
-	cspecialr          csp, mtdc
+	// The thread exit code expects the trusted stack pointer to be in csp and
+	// the stack pointer to be in mtdc.  After thread exit, we don't need the
+	// stack pointer so just put zero there.
+	zeroOne            sp
+	cspecialrw         csp, mtdc, csp
 	j                  .Lthread_exit
 
 	// The continue-resume path expects the location that we will mret to to be