Fix group rule evaluation order with row-level rules

[Copilot]

Group rules were evaluated on the original dataset before row-level filtering. When a row failing a row-level rule would make a group invalid after removal, the group rule passed incorrectly.

class DiagnosisSchema(dy.Schema):
    invoice_id = dy.String(primary_key=True)
    diagnosis = dy.String(primary_key=True, regex="^[A-Z]{3}$")
    is_main = dy.Bool(nullable=False)

    @dy.rule(group_by=["invoice_id"])
    def exactly_one_main_diagnosis(cls) -> pl.Expr:
        return pl.col("is_main").sum() == 1

df = pl.DataFrame({
    "invoice_id": ["A", "A", "A"],
    "diagnosis": ["ABC", "ABD", "123"],  # "123" fails regex
    "is_main": [False, False, True],     # main diagnosis on invalid row
})

# Before: Returns 2 rows with is_main=False (invalid - no main diagnosis)
# After: Returns 0 rows (correctly fails validation)
good, _ = DiagnosisSchema.filter(df)

Changes

• with_evaluation_rules(): Evaluate row-level rules first, then pass to group rule evaluation
• _with_group_rules(): Accept simple_exprs parameter and filter to rows passing all row-level rules before computing group aggregations. Left join results back to original dataset (rows failing row-level rules get null for group rules, treated as valid).
• Tests: Added test_filter_group_rule_after_row_rule_filtering covering interaction scenarios

Group rules now only see data that would remain after row-level filtering.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Validation result is sometimes incorrect when using group rules</issue_title>
<issue_description>In yesterday's PyData meetup in Zurich, one question prompted me to realize that we're incorrectly dealing with group rules and row-level rules: if a row-level rule removes a row which would make a group rule fail, we do not realize it. For example:

import dataframely as dy
import polars as pl

class DiagnosisSchema(dy.Schema):
    invoice_id = dy.String(primary_key=True)
    diagnosis = dy.String(primary_key=True, regex="^[A-Z]{3}$")
    is_main = dy.Bool(nullable=False)

    @dy.rule()
    def exactly_one_main_diagnosis() -> pl.Expr:
        return pl.col("is_main").sum() == 1

df = pl.DataFrame(
    {
        "invoice_id": ["A", "A", "A"],
        "diagnosis": ["ABC", "ABD", "123"],
        "is_main": [False, False, True],
    }
)
good, _ = DiagnosisSchema.filter(df)
print(good)

results in

shape: (2, 3)
┌────────────┬───────────┬─────────┐
│ invoice_id ┆ diagnosis ┆ is_main │
│ ---        ┆ ---       ┆ ---     │
│ str        ┆ str       ┆ bool    │
╞════════════╪═══════════╪═════════╡
│ A          ┆ ABC       ┆ false   │
│ A          ┆ ABD       ┆ false   │
└────────────┴───────────┴─────────┘

which clearly violates the schema since we don't have a main diagnosis for the group.</issue_description>

Comments on the Issue (you are @copilot in this section)

@borchero We should probably fix this by sequentializing group rules and execute them after the row-level rules.

• Fixes Validation result is sometimes incorrect when using group rules #38

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.