Add location field, populate for Sloan courses

[mbertrand]

What are the relevant tickets?

Closes https://github.com/mitodl/hq/issues/5129

Description (What does it do?)

Adds a location field to LearningResource and LearningResourceRun and populates it for Sloan courses (locations for Professional Education courses will be handled after this one is merged in a separate PR). Location isn't relevant for other sources.

The value for LearningResource will be based on the value of the next/best LearningResourceRun.

How can this be tested?

• Copy SEE_ env settings from RC
• Run ./manage.py backpopulate_sloan_data
• Run the following, you should get similar results:

  from learning_resources.models import *
  for location in LearningResource.objects.values_list("location", flat=True).distinct():
      print(location)
  
  'Cambridge, MA', 
  'Melbourne, Australia',
  'Lausanne, Switzerland',
  ...etc

• The search page should still work, before and after running recreate_index --all

Additional Context

[abeglova]

I don't think this is a good data structure. This makes the queries for questions like "List all the courses based in Cambridge, MA" or list "List all the locations which we have upcoming courses" needlessly complicated. Location should be a string field with a single location. Then either learningresourcerun should have an array column with a list of all the location ids for that run or (better) there should be a many to many learningresourcerun_locations table

[abeglova]

It will also make it heard to catch if some locations are duplicates like "Cambridge, MA" vs "Cambridge MA" with no comma - although we don't have a good way of not creating duplicates anyway if Sloan lists the same locations as different strings

[mbertrand]

A learning resource run will only have 1 location. But since different runs may have different locations, I decided to make the learning resource location field an array of all the run values. But I can change it so that it is also a string field equal to the location of the next/best run, similar to availability and next_start_date, and calculated every night. I'm not too keen on a many-to-many relation between courses/runs and a new location model, since location is just a string.

[abeglova]

Can this just be a string array field on learning resource then? I think my objection is having ['Cambridge, MA', 'Lausanne, Switzerland'] and just ['Cambridge, MA'] as seperate location records that multiple courses join to is bad

[abeglova]

Also could ['Lausanne, Switzerland', 'Cambridge, MA'] potentially be created as a separate location record from ['Cambridge, MA', 'Lausanne, Switzerland']

[mbertrand]

Can this just be a string array field on learning resource then

Do you mean "Cambridge, MA; Lausanne, Switzerland" instead of ['Cambridge, MA', 'Lausanne, Switzerland'] for LearningResource.location? I could do that, or just set it to the one location of the next/best run. The location array is sorted in the ETL pipeline so there shouldn't be a ['Lausanne, Switzerland', 'Cambridge, MA'], it should always be ['Cambridge, MA', 'Lausanne, Switzerland'] or 'Cambridge, MA; Lausanne, Switzerland'

I can go ahead and make location a separate model with just a location charfield, with LearningResourceRun having a relation to just one Location object and LearningResource having either the same (identical to that of the next/best run) or a many-to-many relation (all run locations). But it's more joins on queries.

[mbertrand]

Refactored so that location is a CharField for both, and its value for LearningResource is set to the same as the next/best LearningResourceRun in learning_resources.etl.loaders.load_run_dependent_values

[abeglova]

lgtm