.save() ignores schema's data type

[spthm]

It is possible to reassign a value with a different type, e.g.

s = GadgetSnapshot('file_name')
s.load()
hsml = s.hsml[0]
hsml = hsml * np.ones(len(hsml)) * 2.0
s.hsml[0] = hsml

will result s.hsml[0].dtype == dtype('float64'), introducing errors on saving. FortranFile.write_ndarray[s]() should accept a type parameter, and SnapshotBase.save() should pass the type in the schema.

[spthm]

3480d0c somewhat mitigates this by raising an exception on dtype mismatch, but given that is is easy to accidentally assign e.g. a np.float64 array where a np.float32 is required by the schema, automatic conversion would be preferable - at least when it is clearly safe to do so.