tidy + optimize pl$DataFrame, pl$Series

[sorhawell]

summary of refactoring:

• remove internal is_DataFrame_data_input() guard from pl$DataFrame. Reason: not needed, error msg further deep in conversions are better. The guard could also prevent some valid conversion some R class which is not a list or a vector.

• remove pl$DataFrame dead args parallel and via_select. These are not documented and does not do anything, and should have been removed with a previous PR. I would argue this should not count as breaking change to remove some exotic dead args.

• add result_minimal, as_polars_series uses an error to signal internally no impl was found for a given R class or it does not work. result_minimal is light weight version of result which skips all the error upgrading/conversion to signal this.

• raise any other error from as_polars_series which was not due to missing implementation.

• pl$Series drop not needed guard if (inherits(x, "Series")) {... internal conversion will do this also anyways. Remove convert_to_fewer_types() preprocessing which is now covered by as_polars_series.POSIXlt.

This is a follow up to PR #369 that brought optional support for vctrs or any other classed R object via the s3 method as_polars_seri.

When running with debug I see the implementation calls methods left and right and take 1ms! per column. Really only an issue for very wide tables. However, I propose a cleanup of the implementation which also is a bit faster .25ms per small column or so. I have also removed some old code not needed anymore in pl$DataFrame

 > pl$options$debug_polars(TRUE)
 > df = do.call(cbind,lapply(1:100, \(i_) mtcars))
 > dim(df)
 [1]   32 1100
 > system.time(pl$DataFrame(df))
[TIME? ms]
pl$DataFrame() -> [0.797ms]
pl$lit() -> [4.627ms]
  RPolarsErr$plain() -> [0.823ms]
   .pr$RPolarsErr$new() -> [0.159ms]
RPolarsErr$plain() -> [0.7918ms]
.
. {~1100 repeats...}
.
   .pr$RPolarsErr$new() -> [0.149ms]
RPolarsErr$plain() -> [1.21ms]
   .pr$RPolarsErr$new() -> [0.262ms]
RPolarsErr$plain() -> [0.2382ms]
pl$select() -> [0.7279ms]
   .pr$DataFrame$select() ->    user  system elapsed 
  1.205   0.040   1.546 
 
 #with out debug print
> pl$options() -> > pl$options$debug_polars(FALSE)
[1.562ms]
pl$options() -> [0.3068ms]
pl$set_polars_options() -> 
> system.time(pl$DataFrame(df))
   user  system elapsed 
  0.808   0.013   0.825   

and after

> system.time(pl$DataFrame(df))
[9284ms]
pl$DataFrame() -> [9.998ms]
pl$lit() -> [122.3ms]
pl$select() -> [0.5779ms]
   .pr$DataFrame$select() ->    user  system elapsed 
  0.148   0.005   0.149 

Close #336

[etiennebacher]

Cool, always nice to see the internals become less complicated 😄

[etiennebacher]

While you're refactoring pl$DataFrame(), do you want to address #336 as well or leave it for a future PR?

[sorhawell]

it could be some syntactic sugar for $select(pl$col("col1")$cast(new_dtype_1), ... ,... ) that would be ok simple.

If to support direct conversion from any R type to any polars type. That would be a big rework. Since casting is faster than R-polars conversion, it is probably not worth it.

[etiennebacher]

@sorhawell I took a stab at implementing the arg schema in pl$DataFrame(), I hope it's fine to do it here. If you agree with these changes, I can implement it in LazyFrame as well

[etiennebacher]

Actually, I forgot LazyFrame() was just a simple wrapper. I added a few tests and just waiting for CI to pass, thanks @sorhawell!