geff_to_csv handles dtypes poorly

[cmalinmayor]

For me geff_to_csv fails with motile-tracker generated geffs with groups:

Error message:

geff_path_with_groups = "/Users/AStokkermans/Downloads/geff_with_groups.zarr/tracks"

geff_to_csv(geff_path_with_groups, outpath)
---------------------------------------------------------------------------
LossySetitemError                         Traceback (most recent call last)
File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/internals/blocks.py:1175, in Block.putmask(self, mask, new)
   1174 try:
-> 1175     casted = np_can_hold_element(values.dtype, new)
   1177     self = self._maybe_copy(inplace=True)

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/dtypes/cast.py:1830, in np_can_hold_element(dtype, element)
   1829         return element
-> 1830     raise LossySetitemError
   1832 if dtype.kind == "S":
   1833     # TODO: test tests.frame.methods.test_replace tests get here,
   1834     #  need more targeted tests.  xref phofl has a PR about this

LossySetitemError: 

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
Cell In[13], line 1
----> 1 geff_to_csv(geff_path_with_groups, outpath)

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/geff/packages/geff/src/geff/convert/_dataframe.py:112, in geff_to_csv(store, outpath, overwrite)
    109 edge_path = f"{outpath}-edges.csv"
    111 # Convert and write to disk
--> 112 node_df, edge_df = geff_to_dataframes(store)
    113 mode = "w" if overwrite else "x"
    114 node_df.to_csv(node_path, mode=mode)

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/geff/packages/geff/src/geff/convert/_dataframe.py:84, in geff_to_dataframes(store)
     82         series = pd.Series(values)
     83         if missing is not None and any(missing):
---> 84             series.mask(missing, inplace=True)
     85         df_dict[name] = series
     87 dataframes.append(pd.DataFrame(df_dict))

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/generic.py:10491, in NDFrame.mask(self, cond, other, inplace, axis, level)
  10488 if not hasattr(cond, "__invert__"):
  10489     cond = np.array(cond)
> 10491 return self._where(
  10492     ~cond,
  10493     other=other,
  10494     inplace=inplace,
  10495     axis=axis,
  10496     level=level,
  10497 )

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/generic.py:10140, in NDFrame._where(self, cond, other, inplace, axis, level)
  10134     align = self._get_axis_number(axis) == 1
  10136 if inplace:
  10137     # we may have different type blocks come out of putmask, so
  10138     # reconstruct the block manager
> 10140     new_data = self._mgr.putmask(mask=cond, new=other, align=align)
  10141     result = self._constructor_from_mgr(new_data, axes=new_data.axes)
  10142     self._update_inplace(result)

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/internals/managers.py:488, in BaseBlockManager.putmask(self, mask, new, align)
    485     align_keys = ["mask"]
    486     new = extract_array(new, extract_numpy=True)
--> 488 return self.apply(
    489     "putmask",
    490     align_keys=align_keys,
    491     mask=mask,
    492     new=new,
    493 )

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/internals/managers.py:442, in BaseBlockManager.apply(self, f, align_keys, **kwargs)
    440         applied = b.apply(f, **kwargs)
    441     else:
--> 442         applied = getattr(b, f)(**kwargs)
    443     result_blocks = extend_blocks(applied, result_blocks)
    445 out = type(self).from_blocks(result_blocks, [ax.view() for ax in self.axes])

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/internals/blocks.py:1188, in Block.putmask(self, mask, new)
   1183 if self.ndim == 1 or self.shape[0] == 1:
   1184     # no need to split columns
   1186     if not is_list_like(new):
   1187         # using just new[indexer] can't save us the need to cast
-> 1188         return self.coerce_to_target_dtype(
   1189             new, raise_on_upcast=True
   1190         ).putmask(mask, new)
   1191     else:
   1192         indexer = mask.nonzero()[0]

File ~/Documents/Code/HiiragiImageAnalysisGroup/napari_plugins/github/janelia/motile_napari_plugin/.venv/lib/python3.12/site-packages/pandas/core/internals/blocks.py:468, in Block.coerce_to_target_dtype(self, other, raise_on_upcast)
    465     raise_on_upcast = False
    467 if raise_on_upcast:
--> 468     raise TypeError(f"Invalid value '{other}' for dtype '{self.values.dtype}'")
    469 if self.values.dtype == new_dtype:
    470     raise AssertionError(
    471         f"Did not expect new dtype {new_dtype} to equal self.dtype "
    472         f"{self.values.dtype}. Please report a bug at "
    473         "https://github.com/pandas-dev/pandas/issues."
    474     )

TypeError: Invalid value 'nan' for dtype 'bool'

Originally posted by @AnniekStok in https://github.com//pull/407#discussion_r3028540416_

Claude recommended a larger refactor:

It's not just the one bool bug. The series.mask(missing, inplace=True) pattern is fragile across multiple dtypes:

1. bool — the reported bug. NaN can't go into a bool Series.
2. int — pandas silently upcasts to float64 to represent NaN, which loses the integer
   dtype. Not a crash, but a lossy conversion that could break a round-trip.
3. str/bytes — NaN can't go into a plain numpy string/bytes array. Would crash similarly to bool.

And the duplicated masking logic (lines 69-70 and 83-84) means any fix needs to be applied twice, which as you noted is a code smell.