the dipoles file can be reused, provided you are considering the same number of bands. Anyway the timing to calculate them should not depend on the run level.
I do not get this point, if you do an IP calculation the dielectric matrix it is scalar, so you do not have any block size, which is instead taken into account when considering RPA-LF.I found a great difference between the case in which I use an already calculated transition dipole matrix (from IP run), where I can push easily to larger values of the response block size (>3 Ry) and the case in which I do a unique calculation, which is of course longer, but I get memory issues if I don't change the job settings.
2) If you set a box geometry or any other truncation method, the potential does not belong as 1/q^2 anymore so the results are different and it is not just a scaling, anyway the dipoles are not affected by the geometry as they are independent from the coulomb potential.
Best,
Daniele