Track last exited trace (lasttrace) in global_State

The VM now keeps track of the previous vmstate from before the most
recent exit to the interpreter. The field is global_State.lasttrace.

This is intended to help with profiling and diagnostics ("which trace
is to blame for all these expensive exits into the interpreter?")

src/lj_obj.h

@@ -527,6 +527,7 @@ typedef struct global_State {
   GCState gc;		/* Garbage collector. */
   volatile int32_t vmstate;   /* VM state or current JIT code trace number. */
   volatile int32_t gcvmstate; /* Previous VM state (only when state is GC). */
+  volatile int32_t lasttrace; /* VM state before exit to interpreter. */
   SBuf tmpbuf;		/* Temporary string buffer. */
   GCstr strempty;	/* Empty string. */
   uint8_t stremptyz;	/* Zero terminator of empty string. */

src/vm_x64.dasc

@@ -2061,6 +2061,10 @@ static void build_subroutines(BuildCtx *ctx)
   |  mov KBASE, [KBASE+PC2PROTO(k)]
   |  mov L:RB->base, BASE
   |  mov qword [DISPATCH+DISPATCH_GL(jit_base)], 0
+  |  mov TMPRd, dword [DISPATCH+DISPATCH_GL(vmstate)]
+  |  // Record which trace exited to the interpreter, then switch state
+  |  mov TMPRd, dword [DISPATCH+DISPATCH_GL(vmstate)]
+  |  mov dword [DISPATCH+DISPATCH_GL(lasttrace)], TMPRd
   |  set_vmstate INTERP
   |  // Modified copy of ins_next which handles function header dispatch, too.
   |  mov RCd, [PC]