Overhaul availableEntities indexing

[marcelgerber]

For #3388, we need a more accurate representation of the availableEntities field that is currently part of a chart's config.

We want it to

• more accurately match the actual entities that can be found in the chart (not just "overestimating" them)
  • ... by using grapher.tableForSelection
• "default" to the chart tab, which is the more interesting tab usually
• respect the chart's EntitySelectionMode

Implementation & progress

In discussion with @danyx23:

• Create a new table charts_x_entities with columns chartId and entityId (both primary keys, both foreign keys, both indexed)
• Write a script that runs ~weekly and updates this table, using a simple DELETE FROM charts_x_entities; INSERT INTO ... destroy-and-recreate behavior.
• The script instantiates every grapher, downloads its data, looks at grapher.tableForSelection, etc.
• Get rid (-> migration) of the config field data.availableEntities and the code writing it; it is only used for Algolia indexing.
• Think about handling selectedEntityNames/availableEntityNames for chart types like Marimekko, scatter, slope; where currently selection just highlights entities, but leaves unselected entities visible
• The new table charts_x_entities may also be useful for upcoming country pages.

[marcelgerber]

Done in #3418.