Ensure backwards jumps get a yk_location.

I noticed that the mandelbrot benchmark was not triggering the JIT.
This was because we only assigned locations to `for` loops, but this
benchmark contains mostly `while` looping.

Lua has no deidcated bytecode for `while` loops. Instead it uses
unstructured control flow using conditionals and jumps.

Here's an example:
```
i = 10
while i > 0 do
    i = i - 1
end
```

Becomes:
```
main <a.lua:0,0> (11 instructions at 0xbfc160)
0+ params, 2 slots, 1 upvalue, 0 locals, 2 constants, 0 functions
        1       [1]     VARARGPREP      0
        2       [1]     SETTABUP        0 0 1k  ; _ENV "i" 10
        3       [2]     GETTABUP        0 0 0   ; _ENV "i"
        4       [2]     GTI             0 0 0
        5       [2]     JMP             5       ; to 11
        6       [3]     GETTABUP        0 0 0   ; _ENV "i"
        7       [3]     ADDI            0 0 -1
        8       [3]     MMBINI          0 1 7 0 ; __sub
        9       [3]     SETTABUP        0 0 0   ; _ENV "i"
        10      [3]     JMP             -8      ; to 3
        11      [4]     RETURN          0 1 1   ; 0 out
```

In this example `JMP -8` is a CFG back edge that makes the loop.

This commit makes yklua recognise back edges as potential loops so that
they get JITted if/when they get hot. This is of course a proxy: not all
back edges are necessarily loops, but it should (tm) be a decent enough
approximation for our purposes.

I measured about a 12% slowdown with this change. We can't not trace
while loops and expect to be faster, so we will have to take this on for
now.

src/lvm.c

@@ -1222,8 +1222,11 @@ void luaV_execute (lua_State *L, CallInfo *ci) {
     vmfetch();
 #ifdef USE_YK
     YkLocation *ykloc = NULL;
-    if (GET_OPCODE(i) == OP_FORLOOP || GET_OPCODE(i) == OP_TFORLOOP)
+    if ((GET_OPCODE(i) == OP_FORLOOP) ||
+        (GET_OPCODE(i) == OP_TFORLOOP) ||
+        ((GET_OPCODE(i) == OP_JMP) && (GETARG_sJ(i) < 0))) {
       ykloc = &cl->p->yklocs[pcRel(pc, cl->p)];
+    }
     yk_mt_control_point(G(L)->yk_mt, ykloc);
 #endif
     #if 0