There is no unambiguous correct answer to this question because it isn't well posed in terms of logical positivism: what is the difference between the two processes? There is no way to tell which happens if you don't muck up the intermediate state with a measurement.
If you mean this in terms of some quantum field theory with given fields and interactions and asymptotic states, then you can ask how the processes appear in a Feynman description. The scattering process in QED is always two-step, the absorption and emission are separate space-time points. But the emission can precede the absorption both in coordinate time and in proper time along the electron's world line, so you should include "emitted and then absorbed" to the list of possibilities.
Light does not have to be resonant in order to scatter off an atom. The amount of scattering/emission-reabsorption is smaller away from resonance. A light wave is also a long coherent field, and this field can acquire a phase push from the emission-reabsorption, leading the phase-velocity to be bigger than the speed of light.
The issue of "how come the phases add up coherently" is adressed by two things: there is a large scale difference between the atoms and the light wavelength. Each atom scatters the light independently and randomly into a spherical waves, which add up coherently in the original direction only to alter the phase velocity by a constant amount.
There is no scattering from the bulk of a perfect crystal, for long wavelengths, because there is still a discrete translation invariance which means momentum is conserved up to big jumps, and the big jumps give waves with the wrong frequency for long enough wavelengths. But there are discrete momentum additions which are allowed for a short wavelength x-ray in an atomic crystal, and if the photon momentum comes out different but at the same frequency due to the coherence in a different direction, that's called diffraction.
If you want scattering in a crystal, you need to scatter off defects which have a good amount of random variation in a box the size of one wavelength. Similarly, if you scatter of a fluid, you need fluctuations in density to be meaningful in one wavelength. This is easier for blue light than for red light, so transparent fluids scatter blue.