Enable specialized opcode support by default

Summary:
Some microbenchmarks really benefit from this, e.g.
running cinderx/benchmarks/nbody.py with `hyperfine`
looks like:

```lang=plain
# PYTHONJITSPECIALIZEDOPCODES=0
# 15.774s ±  0.239 s

# PYTHONJITSPECIALIZEDOPCODES=1
# 9.630 s ±  0.146 s
```

Reviewed By: DinoV

Differential Revision: D96193032

fbshipit-source-id: 69f2cf8f09816ec39b4d390e58fa89250424de3d

Author: alexmalyshev

cinderx/Jit/config.h

@@ -159,7 +159,7 @@ struct Config {
   bool emit_type_annotation_guards{false};
   // Whether or not to JIT specialized opcodes or to fall back to their generic
   // counterparts.
-  bool specialized_opcodes{false};
+  bool specialized_opcodes{true};
   // Support instrumentation (monitoring/tracing/profiling) by falling back to
   // the interpreter
   bool support_instrumentation{false};