feat: Change mean/median of Date columns to return a Datetime instead of Date

[mcrumiller]

Resolves #13598.

Ref #10232

Part 2 of a series of datetime fixes: implement mean/median for date and return a datetime.

This treats dates and datetimes as analagous to integers/floats: the mean of ints is a float, and now the mean of dates is a datetime:

mean(1, 2) = 1.5
mean(2023-01-01, 2023-01-02) = 2023-01-01 12:00.

Changes:

• return datetime object from mean/median
• mean/median in expr/lit context has pl.Datetime dtype
• .dt.mean() is no longer required for datetime namespace (i.e. has same functionality as s.mean/median).
• group_by(...).mean() now returns pl.Datetime.

Upcoming related PRs:

Part 3 - mean/median for pl.Duration
Part 4 - mean/median for pl.Time
Part 5 - deprecate mean/median for datetime namespace.

[mcrumiller]

@MarcoGorelli mind adding this one to your to-do so I can continue on with the others? This is probably the worst one since I had to update the streaming code a bit to make the date->datetime conversion work in streaming for mean/median.

[stinodego]

I definitely support changing mean and median to return a datetime when appropriate. Is there an issue for this?

[mcrumiller]

@stinodego it was originally issue #10232 with PR #12902 but it was requested to split into multiple PRs. I can make a separate issue for each PR, including this one.

[stinodego]

Thanks - no need to make separate issues I think.

[mcrumiller]

@stinodego oops just made one. It's a bit clearer on the issue at least.

[mcrumiller]

@MarcoGorelli and @reswqa I think I've addressed your issues. Thanks for the review.

[mcrumiller]

@MarcoGorelli and @reswqa I don't want this to get too stale, as I have a few other PRs in the pipeline that are dependent on this one. Do you have any other suggested changes? Thank you!

[reswqa]

Do you have any other suggested changes?

No, I don't. It looks good from my side. But as I am not familiar with temporal thing, I leave the final check before merge to others.

[MarcoGorelli]

Generally looks good to me

The parts I don't want to formally approve myself are:

• crates/polars-pipe/src/executors/sinks/group_by/aggregates/mean.rs
• crates/polars-plan/src/logical_plan/aexpr/schema.rs

but it looks like Weijie is happy with them

[mcrumiller]

Hi @stinodego or @ritchie46 can we merge this so I can proceed with the next temporal updates?

[mcrumiller]

@ritchie46 I hate to harp on this, but it's been sitting in the queue for a while and it's fairly straightforward. Any issues with this PR? I'd like to continue on the other temporal types.

[mcrumiller]

@stinodego I got a streaming error on here, is this related to the markers again?