Owl performance tuning

[gunnar]

The owl pattern VA47a is a very general pattern with a very high value and a non-trivial constraint. Testing to simply remove it gives a balanced breakage with big node reductions:

ld_owl:315      FAIL 0 [1 P18]
strategy:50     PASS Q9 [Q9]
viking:2        FAIL C4 [E2]
strategy2:74    FAIL Q14 [F7|R17|P15]
nicklas2:1402   FAIL H9 [J8|J6]
nngs:990        FAIL L3 [C9]
trevorc:1250    FAIL C2 [!C2]
strategy4:187   PASS E4 [C4|E4]
owl1:293        PASS 0 [0]
nngs4:740       FAIL S5 [D2]
strategy5:295   PASS J2 [J2]
century2002:120 FAIL K2 [L3]
auto04:2        PASS 1 G18 [!0]
ninestones:280  PASS R1 [R1|S1|S2]
manyfaces1:36   FAIL S11 [P16]
seki:1214       FAIL C9 [A1]
cgf2004:30      PASS R6 [P7|R6]
kgs:590         PASS P16 [P16]
olympiad2004:9  PASS B15 [B15]
lazarus:7       pass
strategy2:73    fail
trevor:261      pass
11 PASS (9 PASS, 2 pass)
11 FAIL (10 FAIL, 1 fail)
Total nodes: 1635907282 3111825 12333503 (-2.7% -4.5% -0.26%)

Probably most of the failures can be fixed by replacing VA47a with a small number of more specific patterns.