perf(gen): slow examples generation about ogen

[ernado]

The best solution is to generate already formatted code.

This is actually more annoying to do than waiting for code generation.
We are operating on very big schemas in our example repos, 99.9% users are not affected.

It is pretty general problem for all code generators (and also linters), so I propose:

1. Focus on solving this problem in general
2. Optimize formatting
3. Use concurrency
4. And only then try to generate already formatted code

Anyway, this issue is low priority. IMO, problem with high resource consumption during compilation has bigger impact.

Also: fellows from ent proposed another idea: generate less code (e.g. use reflect-based json over fully code-generated, this can be done with jx too), and same points apply to such optimization.