Fix data race on migrate repository

[lunny]

The data race occupied when both two go routines run repo.getOwner and it will set repo.Owner. Since UpdateRepoIndexer(repo) will send the repo pointer to another goroutine, this PR lets it runs after all other operations finished to resolve the data race. A complete solution to resolve this like problem is to change UpdateRepoIndexer(neededRepoInfo) to only copy needed information from repo.

[lunny]

partially fix #5189

[codecov-io]

Codecov Report

Merging #5224 into master will decrease coverage by 0.07%.
The diff coverage is 57.14%.

@@            Coverage Diff             @@
##           master    #5224      +/-   ##
==========================================
- Coverage   37.54%   37.47%   -0.08%     
==========================================
  Files         310      310              
  Lines       45929    45934       +5     
==========================================
- Hits        17245    17213      -32     
- Misses      26214    26244      +30     
- Partials     2470     2477       +7

Impacted Files	Coverage Δ
models/repo.go	43.55% <57.14%> (-0.02%)	⬇️
models/repo_indexer.go	44.49% <0%> (-10.6%)	⬇️
modules/indexer/repo.go	64.28% <0%> (-5.56%)	⬇️
models/repo_list.go	62.2% <0%> (-1.17%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0ba5794...5c3dfa8. Read the comment docs.

[typeless]

@lunny LGTM.

To prevent such bugs in the future, are the following rules correct (required)?

1. the values of Repository should not be copied.
2. the pointers of Repository should not be passed across the boundary of gourintes.

[lunny]

@typeless I agree with 2, but don't know why we need 1. Copy always cannot result in data race.

[typeless]

@lunny regarding data race, 1. is not required. I made that assumption to be easier to reason about it. I thought the Repository is supposed to be immutable after loaded, but it's not. If I understand correctly, the code updates the Repository objects, and all changes of the objects would be synced back to the database.

If we have two copies of Repository objects for the same git repo and one of them is updated, then the two copies of Repository objects would be inconsistent, which I think would be a problem.