Add streamlit dashboard + refactoring of existing panel dashboard

[csbnw]

Add a Streamlit backend and add the option to switch between the current Panel/Bokeh dashboard and Streamlit/Plotly. Some UI improvements and bug fixes make both dashboards behave consistently. Significant refactoring of the existing code was needed to separate the 'backend' and dashboard-specific code and minimise code duplication.

Notable changes:

• New Streamlit dashboard: Add (streamlit_dashboard.py) using Streamlit/Plotly.
• Multiple backend support: CLI option --backend {panel,streamlit} in ktdashboard.
• Table toggle: Add a UI checkbox to show/hide the raw data. When disabled, the plot fills more vertical space.
• Panel/Bokeh improvements: Make the Bokeh plot responsive (fills page width/height when table hidden).
• Bug fix: Convert categorical values to strings to avoid Bokeh validation errors.
• Separation of backend and dashboard code: Shared logic moved to dashboard.py (parsing, filtering, etc).
• The -demo flag didn't seem to work and is removed.

The panel version now looks like this:

• The plot now fills the page
• An optional data table is shown

And this is the new Streamlit variant:

Note the new option to select a colour palette!

For convenience, when --backend=streamlit is specified, providing a cachefile on the command line is optional. The user can start the dashboard and upload cachefile later:

[benvanwerkhoven]

Wow thanks @csbnw! This looks really nice! I'm looking forward to trying this out. I'll have to familiarize myself with the Streamlit API. I hope it's a better designed than the Panel one, which I wasn't very happy with. I'll have an in depth look at all the code later this week, as I'm currently grading exams.

[stijnh]

Looks great! I tried it and it works very smoothly on existing cache files.

We could also make a new release with only the Streamlit implementation and drop the panel implementation. I'm not sure of having two similar implementations of the same functionality is worth it.