Improvements to ship_traffic

[jbednar]

To-do items for the new https://examples.pyviz.org/ship_traffic example:

• [x] Deploy running server and update website to detect it so that there's a link to it
• [ ] Put another warning near the final app about it only working on a deployed version, and link to the deployed version there.
• [ ] Fix missing axis labels
• [x] Fix thumbnail, preferably by adding one of the colorful categorical plots
• [x] Fix the legend when it's shown by itself
• [ ] Labels are missing from the deployed app if not on localhost, probably because of Stamen restrictions. Is there another tile server for labels we can use instead?
• [ ] Change the two huge PNGs to use separate PNGs per plot so that each plot is more easily cut and pasted and so that the page loads more smoothly. (Also would fix the thumbnail, but really the thumbnail ought to be the more colorful plot anyway.) Could use a HoloMap with a selector by name or a player widget to conserve space on the page.
• [ ] Consider merging tanker and cargo categories to have fewer categorized as unknown; would probably make the Gulf Coast plots more vibrant. Or add a tanker category, which is presumably more important for the Gulf Coast than for the Vancouver region where the category splits were determined.
• [ ] Select an initial location for the app selection that doesn't happen to match a different category than what's visible (tanker vs cargo).
• [x] Speed up the deployed app, at least for the second visitor; not sure what's slowing deployment down
• [ ] Update to use https://github.com/holoviz/holoviews/pull/4806 once released (also needed for census, nyc_buildings, maybe some others)
• [ ] Make the reset tool for the app reset to the full range of data available, even if it starts zoomed in
• [ ] Break up loading code where possible so that it's more clearly separated between the cached version and the version building the cache.
• [ ] Add a template to format the final app nicely
• [ ] Either really exploit spatial indexing, or remove it to keep it simpler and more direct (just a Pandas or Dask df, with indexing only when actually used)
• [ ] Shorten the URL field shown in the table; ideally the MMSI would just be a clickable link with no separate URL field
• [ ] Isn't there a better way to construct a color key than {list(groups.keys())[i]:tuple(int(e*255.) for e in v) for i,v in enumerate(colors[:(len(groups))][::-1])}? Seems awkward.