Narrow aggregation exception guards to TypeError

[vahid-ahmadi]

Closes #304

Problem

The aggregation wrappers in microdataframe.py wrapped each per-column call in a bare except Exception: pass — eight occurrences across _create_scalar_function, _create_vector_function, _create_agnostic_function, and the two MicroDataFrameGroupBy factories.

The intent (#213) is to skip columns that can't be aggregated. But catching everything also discarded genuine errors and handed back a silently truncated result:

d = MicroDataFrame(pd.DataFrame({"x": [1, 2, 3]}), weights=[1, 1, 1])
d.gini(negatives="bogus")   # -> Series([], dtype: object)

MicroSeries.gini raises a deliberate, well-worded ValueError for an unknown negatives option (#289). At the DataFrame level it vanished — no traceback, no warning, nothing to say the argument was wrong. The same guard would hide a genuine bug in any weighted statistic.

Change

• Catch TypeError only — the exception a dtype mismatch actually raises.
• Log the skip via logger.debug("skipping column %s in %s: %s", ...) so a swallowed failure is discoverable.

Which columns aggregate does not change: non-numeric columns are already filtered by the is_numeric_dtype check above each try. Verified against a frame mixing int, str, bool, and datetime columns — the same columns come back before and after (sum, mean, median, gini, quantile, decile_rank, and groupby(...).sum()).

Verification

• Two new tests: the ValueError now propagates, and non-numeric columns are still skipped
• make test — 64 passed on pandas 3.0.5 and pandas 2.3.3

Note: the lint job will be red until #307 merges (unrelated ruff 0.16 issue on main).