Fix custom function input not binding positionally when column name d…

[mborodii-prog]

Summary

Fixes #747 — custom row-wise functions silently dropped a column's value whenever it didn't exactly match one of the function's parameter names, producing a confusing TypeError instead of resolving the function's other parameters against their defaults.

• Row-wise custom functions previously bound columns to parameters only by exact name equality (or via **kwargs). If a column's name didn't match a parameter name, its value was discarded entirely and the function was called without it.
• Now, when none of the available column(s) match a parameter by name they fall back to binding positionally, in the function's declared parameter order, to whichever parameters aren't already supplied explicitly — whether the columns came from an explicit input: or from the dataframe as a whole when input: isn't given at all.
• When there are more unmatched columns than the function has room for, it's ambiguous which ones to use — instead of guessing (or falling through to a misleading "missing argument" error), this now raises a clear ValueError naming exactly what's expected vs. what was given.

Examples

def func(x, y="default", z="default"):
    return f"{x}-{y}-{z}"

1. Mismatched column name via input:

wrangles:
  - custom.func:
      input: my_col
      output: result

	Before	After
Column my_col → param x, y/z default	TypeError: func() missing 1 required positional argument: 'x'	result = "row1-default-default"

2. Mismatched column, no input: at all

wrangles:
  - custom.func:
      output: result

Dataframe has one column my_col (no name match) → same result as above,
result = "row1-default-default". Previously this also failed the same way.

3. Multiple mismatched columns bind positionally, in order

wrangles:
  - custom.func:
      output: result

Columns present	Binding	result
col_a	col_a → x; y/z default	"col_a_v1-default-default"
col_a, col_b	col_a → x, col_b → y; z default	"col_a_v1-col_b_v1-default"
col_a, col_b, col_c	col_a → x, col_b → y, col_c → z	"col_a_v1-col_b_v1-col_c_v1"

4. Too many unmatched columns → clear error

wrangles:
  - custom.func:
      output: result

Dataframe has 4 columns (a, b, c, d) for a 3-parameter function:

	Before	After
4 columns, 3 params	TypeError: func() missing 1 required positional argument: 'x' (misleading — implies something's missing, not that too much was given)	ValueError: custom.func accepts at most 3 unfilled parameter(s) (x, y, z) but 4 column(s) were provided: a, b, c, d

The fallback only ever engages when there's zero name-based match, so existing name-matching and literal input/output-as-parameter behavior is unchanged.

Changes

• wrangles/recipe.py: positional-binding fallback and too-many-columns guard in the row-wise custom-function dispatch (_execute_wrangles).
• tests/recipes/test_custom_functions.py: regression tests for each of the four scenarios above.
• tests/recipes/wrangles/test_extract.py: unrelated flaky-test fix — test_ai_invalid_model_per_row_error was pinned to status=400 for an unknown model, but OpenAI now reports that as 404 model_not_found; relaxed to accept any 4xx client error.