fix(with_columns): require exactly one of pass_dataframe_as / columns_to_pass / on_input

[Hore01]

Summary

with_columns_base.__init__ is meant to enforce mutual exclusivity between pass_dataframe_as, columns_to_pass, and on_input — the docstring on each of those parameters says so. The current check raises only when exactly two are set:

if (
    int(pass_dataframe_as is None) + int(columns_to_pass is None) + int(on_input is None)
    == 1
):
    raise ValueError("You must specify only one of ...")

So a caller who specifies all three silently passes the validation and crashes later with a confusing downstream error. A caller who specifies zero of them also silently passes.

Args set	sum(is None)	currently raises?	should raise?
0	3	no	yes
1	2	no	no
2	1	yes	yes
3	0	no	yes

This PR replaces the boolean arithmetic with the intuitive form:

n_set = sum(arg is not None for arg in (pass_dataframe_as, columns_to_pass, on_input))
if n_set != 1:
    raise ValueError("You must specify exactly one of ...")

The error message opener is also tightened: "only one of" → "exactly one of".

Tests

Four new tests in tests/function_modifiers/test_recursive.py, exercising the base class directly via a minimal inline subclass:

• 0 args set → ValueError
• 2 args set → ValueError (regression of the case that was already caught)
• 3 args set → ValueError (new — was the primary bug)
• 1 arg set → no error (positive case across all three options)

Validation lives in the shared base class so no parallel h_polars / h_spark tests are necessary. Testing against with_columns_base directly (rather than e.g. h_pandas.with_columns) is necessary because some subclasses pre-reject certain arg combinations before delegating to the base.

Behaviour change

Callers passing 0 or 3 of the mutually-exclusive args used to silently pass the decoration step and crash later with confusing errors deep in the DAG-build path. After this PR they get a clean ValueError at decoration time. This is the only intentional behavioural change. The documented contract has always been mutual exclusivity, so no caller should be relying on the silent-pass behaviour — but flagging the change explicitly so reviewers aren't surprised.

Signed-off-by: Olajumoke Akinremi 106763970+Hore01@users.noreply.github.com