okay, I find the explanation finally. Since magnesium halides are moderate Lewis acids,their presence in solution may influence the outcome of certain chemical reactions. Magnesium bromide rearranges the epoxide to cyclopentanecarbaldehyde, which then adds the Grignard reagent in the expected manner like your drawing. dialkylmagnesium reagents and organolithium reagents simply adds to the epoxide by opening the strained ring, they are more selective. For the mechanism of the rearrangement, I am not sure, I think it may like the pinacco rearrangement.