Compute a rolling mean (window 3) within each group, ignoring NaN values. df has columns 'group' and 'value'. Return the resulting Series aligned to the original index.

Sample data: df:

  group  value
0     a    1.0
1     a    2.0
2     b    3.0
3     b    NaN
4     a    2.0
5     b    3.0
... (12 rows total)