Include new instruction execute_cpred to perform tail optimisation for

builtins. Required changes:
- be careful about creeping in deallocate: it may be followed by
something that is not a cut nor a proceed.
- include new instruction in absmi.c: it is a merge of execute and
call_cpred.
- change compiler to generate execute even for C builtins.
- be careful with dexecute: it may not be done if execute_op is a C
builtin.
- if we are in execute_cpred, the garbage collector cannot trust P:
instead it must look at CP to find out the size of the current
environment. The macro gc_P receives that information.
- We don't need to change CP if we do a meta-call from within
execute_cpred (and we in fact cannot). Check places where we do
meta-calls: exec, clause in cdmgr, and lu_recorded.

C/index.c

@@ -944,6 +944,7 @@ has_cut(yamop *pc)
       break;
     case _execute:
     case _dexecute:
+    case _p_execute_cpred:
       pc = NEXTOP(pc,pp);
       break;
       /* instructions type l */
@@ -2467,6 +2468,7 @@ add_info(ClauseDef *clause, UInt regno)
     case _retry_profiled:
     case _count_retry:
     case _execute:
+    case _execute_cpred:
     case _dexecute:
     case _jump:
     case _move_back: