op_scope/newCONDOP: Optimise away empty else blocks in OP_COND_EXPR

[richardleach]

This commit comprises two main changes:

• Not wrapping a bare OP_STUB in an enter/leave pair (Perl_op_scope)
• Checking for a bare OP_STUB in the falseop argument when building an OP_COND_EXPR and then freeing it, rather than adding it as a sibling to the first and trueop branches.

The main benefits of this are:

• lower memory usage due to unnecessary OP removal
• faster execution due to unnecessary OPs not being executed

There are also some small changes to Deparse.pm and an additional Deparse.t test.

---

• This set of changes requires a perldelta entry, but we are close enough to a point release that I'll add it separately.

[richardleach]

As a quick example, my $x; if ($x) { say 1 } else { }:
Before:

a  <@> leave[1 ref] vKP/REFC ->(end)
1     <0> enter v ->2
2     <;> nextstate(main 2 -e:1) v:%,us,{,fea=15 ->3
3     <0> padsv[$x:2,9] vM/LVINTRO ->4
4     <;> nextstate(main 3 -e:1) v:%,us,{,fea=15 ->5
-     <1> null vK/1 ->a
6        <|> cond_expr(other->7) vK/1 ->b
5           <0> padsv[$x:2,9] s ->6
-           <@> scope vK ->-
-              <;> ex-nextstate(main 5 -e:1) v:%,us,fea=15 ->7
9              <@> say vK ->a
7                 <0> pushmark s ->8
8                 <$> const[IV 1] s ->9
d           <@> leave vK ->a
b              <0> enter v ->c
c              <0> stub vP ->d

Now:

a  <@> leave[1 ref] vKP/REFC ->(end)
1     <0> enter v ->2
2     <;> nextstate(main 8 -e:1) v:%,us,{,fea=15 ->3
3     <0> padsv[$x:8,15] vM/LVINTRO ->4
4     <;> nextstate(main 9 -e:1) v:%,us,{,fea=15 ->5
-     <1> null vK/1 ->a
6        <|> cond_expr(other->7) vK/1 ->a
5           <0> padsv[$x:8,15] s ->6
-           <@> scope vK ->-
-              <;> ex-nextstate(main 11 -e:1) v:%,us,fea=15 ->7
9              <@> say vK ->a
7                 <0> pushmark s ->8
8                 <$> const(IV 1) s ->9

[richardleach]

The op.c changes to omit an empty if-true are also straightforward, but Deparse looks like being more work. This is why there isn't a single PR to cover both.