Op re-entrancy guard is too wide

[mmastrac]

Ideally the re-entrancy guard should only protect the op call itself, not the surrounding error-handling code that is out of control of the op writer.

[mlafeldt]

I was actually wondering about this when I saw the error message caused by a custom op invoking the op_is_native_error built-in.

Unexpectedly panicked!: op op_sign_ecdsa was not marked as #[op2(reentrant)], but re-entrantly invoked op op_is_native_error