Expression output field must match engine output

[kdn36]

Some expressions set a different output field from the one produced by the query engine. In other words: LazyFrame.collect_schema() must match LazyFrame.collect().schema for any expression. Each mismatch is technically a bug, and may impact execution (depending on the execution context).

Initial scope: ApplyExpr.

Detailed scope: see the mismatch baseline at #23649.

Known problematic cases:

• str.to_integer (fix: Fix output field dtype for ToInteger #23664)
• business_day_count to use LHS-convention for output field name (fix: Match output field name to lhs for BusinessDaycount #23679)
• cum_sum_horizontal (fix: Match output dtype to engine for cum_sum_horizontal #23686 )
• get_arithmetic_field with struct and numeric
• str.extract_groups with the empty string pattern (fix: Fix output for str.extract_groups with empty string pattern #23698)
• rolling_map (fix: Match output type to engine for rolling_map #23702)
• interpolate() on pl.Decimal (fix: Match output type to engine for interpolate on Decimal #23706)
• over(pl.struct("x"))
• when_then_over
• str.to_decimal
  Scale is resolved based on data. Should get a scale parameter and return unknown if not given.
• str.to_datetime / str.strptime(pl.DateTime)
  Resolves time zone based on data if format is not given. Time unit is don't resolved based on data, but has some complex rules to it.
• shrink_dtype
  Output datatype is resolved based on data. Should return unknown.
• when_then Output type mismatch in when-then #23733

[thoo]

• when_then : added PR here. fix(rust): Fix schema consistency for Categorical in when-then expressions #23762