fix(backup): Fix mis-migrated AlertRule imports

[azaslavsky]

In theory, as part of the ongoing "actor refactor", all AlertRule instances with an owner_id pointing to a non-null Actor should have that Actor instance's populated user_id or team_id field hoisted into the AlertRule model (see #58667 for more). In practice, we've seen some failing relocations of real self-hosted data, implying that there have been some misses in the migration, causing newly imported models to fail validation.

In the long term, we'd like to implement a migration that fixes all of the missing edge cases. Until that happens, this import-time fix should allow us to work around the issue. The basic logic is to manually do the hoisting at import-time if possible. If not possible (either due to a missing owner_id, or because the Actor being pointed to both team and user set to null), we ensure that the owner_id is nulled out.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (bf24a1a) 81.25% compared to head (20dab6c) 81.25%.
Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master   #62526   +/-   ##
=======================================
  Coverage   81.25%   81.25%           
=======================================
  Files        5197     5197           
  Lines      230098   230110   +12     
  Branches    39745    39748    +3     
=======================================
+ Hits       186960   186975   +15     
+ Misses      37399    37398    -1     
+ Partials     5739     5737    -2     

Files	Coverage Δ
src/sentry/incidents/models.py	96.62% <100.00%> (+0.08%)	⬆️
static/app/components/events/eventReplay/index.tsx	71.42% <ø> (ø)

... and 5 files with indirect coverage changes