Fix sum() axis functionality

[juaristi22]

Fix #262

[MaxGhenis]

Heads-up: I opened #283 to fix a related-but-distinct bug — weighted sum() silently returns 0.0 on a MicroSeries/MicroDataFrame with a non-default index (because set_weights stored the weights Series with a default RangeIndex). The fix touches microseries.py set_weights and the MicroDataFrame set_weights, so the diff shouldn't conflict with #263 (which is about sum(axis=...)). Flagging in case of merge-conflict-detective-work later.

[vahid-ahmadi]

Good fix, and it quietly repairs more than the title suggests: before this, mdf.sum(axis=0) also returned an empty Series, because axis=0 was forwarded into MicroSeries.sum() (which takes no arguments), raised TypeError, and got swallowed by the except Exception: pass. Stripping axis out of ms_kwargs fixes that too. The test coverage on skipna/min_count/axis="columns" is thorough.

One real gap and a couple of smaller notes.

1. Positional axis is still broken — df.sum(1) silently returns an empty Series.

axis = kwargs.get("axis", 0)

pandas fully supports df.sum(1), and it's a common shorthand. With a positional argument, kwargs is empty, so axis reads as 0, the column-wise branch runs, and getattr(self[col], "sum")(1) raises TypeError against MicroSeries.sum(self) — caught by except Exception: pass, every column dropped, empty Series returned. Same failure mode you're fixing, just via the other calling convention.

Worth pulling axis out of the positional slot:

def fn(*args, **kwargs):
    if args and "axis" not in kwargs:
        axis, *rest = args
        args = tuple(rest)
    else:
        axis = kwargs.get("axis", 0)
    ms_kwargs = {k: v for k, v in kwargs.items() if k != "axis"}

That also removes a latent problem in the axis=1 branch: getattr(df, name)(axis=1, *args, **ms_kwargs) is legal Python but raises TypeError: got multiple values for argument 'axis' the moment args is non-empty, since axis is pandas' first positional parameter.

2. axis=1 reaches functions pandas doesn't have.

SCALAR_FUNCTIONS is derived from every MicroSeries method with _rtype == float (microseries.py:821-826), which includes gini, top_10_pct_share, top_x_pct_share, t10_b50, and friends. getattr(pd.DataFrame, "gini") doesn't exist, so mdf.gini(axis=1) now raises a bare AttributeError naming DataFrame — confusing, given the user called a method on a MicroDataFrame. A row-wise gini is meaningless anyway, so an explicit refusal would read better:

if not hasattr(pd.DataFrame, name):
    raise TypeError(f"{name}() does not support axis=1 on a MicroDataFrame.")

3. Minor

• The docstring on _create_scalar_function was dropped in the rewrite; the surrounding methods (_create_vector_function, _create_agnostic_function) still carry theirs.
• The return annotation is Union[pd.Series, float], but no path returns a float.
• axis=None (explicit) falls into the else and raises ValueError, where pandas treats it as "reduce over everything". Probably fine to leave, just noting it's now a hard error rather than the old silent-empty.

Also — GitHub currently shows this as conflicting with main, so it'll need a rebase before it can go in.