Improvements and Documentation for Panes

[philippjfr]

This PR provides polish to all existing pane classes and then adds reference sections for each. I'll outline all the panes and the improvements made here:

• HoloViews
  • Add support for widget state overrides as dictionary in widgets parameter (fixes https://github.com/pyviz/panel/issues/302)
  • Ensure that initial plot state matches widget state
  • Add reference gallery entry
• Plotly
  • Fixed and improved efficiency of updates when only data or only layout properties are changed
  • Ensure it is rendered correctly on initialization
  • Fixes issues sending arrays of different shapes (fixes ColumnDataSource's columns must be of the same length.  #226)
  • Add reference gallery entry
• Vega
  • Ensure it is rendered correctly on initialization
  • Add reference gallery entry

[codecov-io]

Codecov Report

Merging #313 into master will decrease coverage by 0.09%.
The diff coverage is 75.75%.

@@           Coverage Diff            @@
##           master    #313     +/-   ##
========================================
- Coverage   89.29%   89.2%   -0.1%     
========================================
  Files          67      67             
  Lines        6622    6696     +74     
========================================
+ Hits         5913    5973     +60     
- Misses        709     723     +14

Impacted Files	Coverage Δ
panel/tests/test_plotly.py	97.46% <100%> (ø)	⬆️
panel/models/plotly.py	100% <100%> (ø)	⬆️
panel/models/mathjax.py	100% <100%> (ø)	⬆️
panel/models/katex.py	100% <100%> (ø)	⬆️
panel/models/vega.py	100% <100%> (ø)	⬆️
panel/tests/test_holoviews.py	99.25% <100%> (+0.03%)	⬆️
panel/pane/vega.py	78.37% <100%> (ø)	⬆️
panel/io.py	63.69% <33.33%> (-3.09%)	⬇️
panel/pane/plotly.py	84.44% <80.64%> (-2.52%)	⬇️
panel/pane/holoviews.py	88.99% <87.5%> (+5.1%)	⬆️
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 16f32ea...70c106e. Read the comment docs.

[philippjfr]

So after refactoring the various models I now realize that loading the JS in this does not work for the classic notebook as noted in bokeh/bokeh#5227.

Not sure what to do about this yet, but it may require a fallback condition implemented using jQuery or similar.

[philippjfr]

In the worst case scenario which is that we find no non-hacky solution that could be added to bokeh itself I will likely just insert this hack in our JS:

if (window.requirejs) {
  window._define = window.define;
  // Stub does not define amd and is therefore skipped by UMD loaders
  // however Notebook bundle does not check for window.define.amd
  function _define_stub(name, deps, callback) {
    window._define(name, deps, callback)
  }
  window.define = _define_stub;
  function _reset_define() {
    window.define = window._define;
  }
  window._bokeh_onload_callbacks.push(_reset_define)
}

It temporarily replaces window.define with a stub which is not recognizable by a UMD loader but will still be used to fetch the classic notebook webpack bundle.

[philippjfr]

I decided to trial a fix for the issue loading dependencies in classic notebooks by adding a __js_require__ attribute to custom models, which is used in the autoload_panel_js.js template to generate a requirejs call which satisfies the dependencies. If this works well for us I'll fully flesh out the __js_require__ API and contribute it upstream.

@xavArtley If you want the VTK model to work in the classic notebook you'll be able to add something like this to your VTK model to load vtk.js correctly:

__js_require__ = {"paths": {"vtk": "//unpkg.com/vtk.js@8.3.15/dist/vtk"},
                  "shim": {"vtk": {"exports": "vtk"}}}