It is expected that for most VI's, you will use surface reflectance. If there are little atmospheric effects (such as with low flying UAV data), then the TOA and surface reflectance can be very similar. However, we do mention in the documentation for VIs that some areas of low radiance/reflectance, such as shadow or water, can be unreliable estimates for VI conditions:

"Applying VIs to vegetative materials to analyze their conditions does not produce absolute results. The results derived from the VIs are relative when determining where distribution or stress is, providing a theoretical level ranging from a low to high value.
Shadowed or highly shaded areas in an input file may not have enough light to cause the vegetation signal to register for the VIs. Atmospheric correction does not improve the quality of these areas. Under these conditions, the VIs may provide inaccurate representations of the vegetation conditions."

VI is Vegetation Index's? ... but what for Water Index? NDWI is the same consideration?

Yes, NDWI is still considered a vegetation index because it is looking at vegetation canopy water. Here is the definition:
This index is sensitive to changes in vegetation canopy water content because reflectance at 857 nm and 1241 nm has similar but slightly different liquid water absorption properties. The scattering of light by vegetation canopies enhances the weak liquid water absorption at 1241 nm. Applications include forest canopy stress analysis, leaf area index studies in densely foliated vegetation, plant productivity modeling, and fire susceptibility studies.