delay costly makeError to when it is needed

[mgilbir]

In the context of issue #205, this change makes parsing 4x faster in my tests.

[colinhacks]

Can you summarize what you did here? Everything was prettified with a shorter printWidth so it's hard to find your changes.

[colinhacks]

Ah I just saw you're other comment. Basically your moving the makeError definition out of the parsing function. Makes sense! I'll play around with this.

[mgilbir]

Agh, sorry! I only saw now how messy the diff got due to applying prettier. I was rushing to send this between meetings and I didn't double check the diff before pushing 😞

I moved the makeError out fo the parsing because in the profiler I saw that the ZodError generation was always present and taking a significant amount of time on JSON blobs that never error. Doing less work on a heavy-hit function was bound to speed things up.

I didn't see any other obvious things in the profiler, so I didn't dig deeper.

[colinhacks]

Merged into Zod v1 and v2-beta. Thanks @mgilbir, great work on this!