The proof-theoretic ordinal of first-order arithmetic ($\mathsf{PA}$) is $\varepsilon_{0}$. However, in pages 3 and 4 of Andreas Weiermann's Analytic combinatorics, proof-theoretic ordinals, and phase transitions for independence results, he mentions it is possible to define a primitive recursive well-ordering $\prec$ whose order type is $\omega$ but such that $\mathsf{PA}$ cannot prove transfinite induction along $\omega$. No reference is provided other than the name Kreisel.
First, does anyone know how to construct $\prec$, or perhaps may provide a modern reference where this is accomplished?
Second, if such a pathological well-ordering $\prec$ exists, could not we use it to prove the consistency of $\mathsf{PA}$ by using finitary methods plus transfinite induction along $\prec$? If this is the case, why do we define the proof-theoretic ordinal of a theory $T$ the way we do, and not simply as
The least ordinal $\alpha$ such that transfinite induction along $\alpha$ (plus finitary methods) proves the consistency of $T$.
I guess this would depend on the ordinal notation used; and since the definition of proof-theoretic ordinal does not depend on a notation, it is then more robust. Is that it, or there is more?
EDIT: I just found out that my question is also answered in Math Overflow. There are very informative and useful answers provided there.