Fix segfault in groupby-var

Previously we'd accidently call compute here when inferring metadata.
The new groupby implementation segfaults when threads are used to
compute the index (apparently), which could happen in some cases.  It's
easier just to let `aca` handle the metadata inference, so we'll remove
it.

dask/dataframe/groupby.py

@@ -251,14 +251,10 @@ def mean(self):
 
     @derived_from(pd.core.groupby.GroupBy)
     def var(self, ddof=1):
-        from functools import partial
-        meta = self.obj._meta
-        if isinstance(meta, pd.Series):
-            meta = meta.to_frame()
-        meta = meta.groupby(self.index).var(ddof=1)
         result = aca([self.obj, self.index], chunk=_var_chunk,
-                     aggregate=partial(_var_agg, ddof=ddof), meta=meta,
-                     token=self._token_prefix + 'var')
+                     aggregate=_var_agg,
+                     token=self._token_prefix + 'var',
+                     aggregate_kwargs={'ddof': ddof})
 
         if isinstance(self.obj, Series):
             result = result[result.columns[0]]