Since I have never really liked Matplotlib I would really like to learn Bokeh, but unfortunately its inability to export its visualizations in SVG or similar formats makes it kind of useless for me as a scientist wanting to publish my results.
to_bokeh(fig=None, use_pandas=True, xkcd=False)
That being said, this sounds nice for adding interactivity (sliders to vary parameters, that sort of stuff). Useful in presentations and the sort.
At the current time you can export to png, or maybe export to svg in a roundabout manner, by turning html page into pdf, turning pdf into svg, etc.
Raster does have limited resolution, but so does your screen or most output devices (yes, even scientific plotters).
It's also not necessarily true that it has a higher storage footprint; it depends on the complexity and number of glyphs, the fonts you need to embed for the LaTeX formulae, etc. etc. In large-data cases, raster can be a much more viable visualization transfer medium.
SVG is nice too, just needs few more intermediate steps.