Switch cohort loading process to celery rather than request thread

[jonespm]

The cohort loading process can take awhile and we've had to implement fixes to work around this like increasing timeouts TLPSO-465. A future version of this tool should pass this work off to a backend process to do the load rather than having the user wait for it.

[jonespm]

I also noticed when investigating the cohort loading process that the actual process is really inefficient.

• For every row in the Excel file 1) select from management_student and management_mentor for every user to see if they exist
  • if they haven't add that record 2) select from management_studentcohortmentor for every row to see if they exist
  • if they haven't add that record

None of this is batched, it's all single queries. There's no reason for the extra individual selects in step ii. since this doesn't actually function as an update. This would be more efficient if the selects in i. for mentor and student were from the inserts, the inserts in i. were bulk for those missing, then ii. were inserted in