smarter, faster, more memory-efficient, lazy pivot

[jqnatividad]

Even though we load the CSV in lazy mode (primarily to faciltate date inferencing with --try-parsedates and --infer-len), pivotp currently uses Polars' eager mode to do the pivot.

https://docs.pola.rs/user-guide/concepts/lazy-api/

https://github.com/pola-rs/polars/blob/9ea5839c52bf0606aaa0b174d9a974992e0ea328/crates/polars-ops/src/frame/pivot/mod.rs#L119-L153

Since we already use the stats-cache to do pivot validation and smart aggregation, use the stats cache to infer a polars schema as well and do the pivot using the Lazy API's group by method instead.