geolocation information of Landsat products is usually very good, especially images from newly released "Collection" (https://landsat.usgs.gov/landsat-collections).

Furthermore, at areas with low topography (unlike mountainous regions) georeferencing should be sufficient.

Co-registering for multi-temporal analyses is basically a good idea. In your case I would say that you could probably introduce erros because changes in sea level could alter the shoreline (which you are exactly interested in), but co-registration could wrongly interpret this as misregistered parts of the image and 'correct' these areas resulting from temporal changes. You would then lose the information on shoreline variation.

Manually setting single GCPs however at places of constant location could help getting higher accuracies.

Thanks Andreas!