Save Ri, Rj and appropriate transposes in matrices for efficient computation in PreintegrationBase.h

[krunal chande]

Currently they are being assigned to Rot3, which is expensive as we need the matrix version often. Plus, if we use quaternions, it would waste lot of time for the conversion to Rot3.

[Frank Dellaert]

Merged in fix/ImuFactorInefficiencies (pull request #133)

Small optimization improvements in PreintegrationBase.h. Fixes Issue#227

→ <<cset eb8d1a91a60e>>