Fix a crash bug when GC is ran while creating a proc with env

mrb_proc_new_cfunc_with_env() allocates RProc with RProc::env as NULL
then allocates REnv and sets it to RProc::env of the allocated RProc. If
incremental GC is ran before "allocates REnv and sets it to RProc::env
of the allocated RProc", the allocated RProc's GC status is
"marked" (Black) and the allocated REnv's GC status is
"unmarked" (White). The next incremental GC sweeps the allocated REnv
without re-marking the allocated RProc. Because the RProc is Black and
the REnv is White.

We need to implement write barrier for the case.

We can force to cause the above situation by the following patch:

    diff --git a/src/proc.c b/src/proc.c
    index f98998f..4f4e25c 100644
    --- a/src/proc.c
    +++ b/src/proc.c
    @@ -92,6 +92,7 @@ mrb_proc_new_cfunc_with_env(mrb_state *mrb, mrb_func_t func, mrb_int argc, const
       struct REnv *e;
       int i;

    +  mrb_incremental_gc(mrb);
       p->env = e = env_new(mrb, argc);
       MRB_ENV_UNSHARE_STACK(e);
       e->stack = (mrb_value*)mrb_malloc(mrb, sizeof(mrb_value) * argc);

With this patch, "rake test" causes segmentation fault.

src/proc.c

@@ -93,6 +93,7 @@ mrb_proc_new_cfunc_with_env(mrb_state *mrb, mrb_func_t func, mrb_int argc, const
   int i;
 
   p->env = e = env_new(mrb, argc);
+  mrb_field_write_barrier(mrb, (struct RBasic *)p, (struct RBasic *)p->env);
   MRB_ENV_UNSHARE_STACK(e);
   e->stack = (mrb_value*)mrb_malloc(mrb, sizeof(mrb_value) * argc);
   if (argv) {