feat: support column-projected worksheet reads

[alexander-beedie]

Ref: pola-rs/polars#27677

This PR allows for projecting columns at read time (for xlsx/xlsb), so we can omit columns from the streaming reader before collection - this means peak allocation tracks the selected columns, not the entire sheet's Range.

The linked issue shows a ~40GB allocation which would only need to be a few MB if the required columns could be declared (and the remaining columns filtered out) before we allocate.

[lukapeschke]

Thanks for addressing this! I'm thinking we could maybe make the new methods take an non-exhaustive options struct ? This would allow to add support for more options later without having to modify the existing API.

Something like this:

#[derive(Debug, Clone, Default)]
#[non_exhaustive]
pub struct RangeOptions {
    column_indices: Option<Vec<u32>>, // or an ordered set. Or (u32, u32) coordinates
    // In another PR, allow to select rows as well
    row_indices: Option<Vec<u32>>, 
}


fn worksheet_range_opts(
       &mut self,
       name: &str,
       opts: RangeOptions,
   ) -> Result<Range<Data>, Self::Error> {]

This would allow us to select only the rows we need , and we could also move the header_row option there eventually, since it's more sheet-specific than reader-specific.

Looks good otherwise, thanks for the extra tests 👍

[alexander-beedie]

Thanks for addressing this! I'm thinking we could maybe make the new methods take an non-exhaustive options struct ? This would allow to add support for more options later without having to modify the existing API.

@lukapeschke: adding support for rows here is definitely a good move; do you have an idea if there are other meaningful options that would be useful outside of row/col definitions?

In the meantime I'll drop this PR into Draft and start adding rows support right away. I also realised it would be nice to accept RangeBounds for improved ergonomics - probably just need suitable From/Into definitions...

reader.worksheet_range_ref_cols("Sheet1", 0..5)?;       // range
reader.worksheet_range_ref_cols("Sheet1", 0..=4)?;      // inclusive
reader.worksheet_range_ref_cols("Sheet1", 2..)?;        // from col 2 to end
reader.worksheet_range_ref_cols("Sheet1", [1, 3, 5])?;  // explicit/disjoint list
reader.worksheet_range_ref_cols("Sheet1", ..)?;         // all (no projection)

...but ideally we would support declaring disjoint ranges/indices in one go to select a region 🤔

reader.worksheet_range_ref_region("S", 0..5, ..)?;           // range of cols, all rows
reader.worksheet_range_ref_region("S", [1,3,5], 0..100)?;    // specific cols, range of rows
reader.worksheet_range_ref_region("S", [0..3, 8..10], ..)?;  // disjoint col ranges, all rows

[alexander-beedie]

Closing here and raising as a fresh PR... New version now handles regions (that can potentially be disjoint) instead of only cols, and is essentially a rewrite 👌