Change to use pre-trained model feature to manage REconverge and Hog models

[syncpark]

Issues:

1. Currently REconverge model is stored in Postgres DB. Problems occur when saving the model file if it is too large.
2. Hog models are not maintained currently.

Proposal:

1. Let's store all model files into pretrained directory instead of Postgres DB.
2. Change model table
   • Add description column to store model related information
   • Add path to specify model file path
3. Add or change GraphQL APIs
   • Add new APIs to register or remove a model
   • Change models API to query model list including kind, description
4. Extend model structure to store other kind of models like Hog's models

[syncpark]

@kimhanbeom @MW-Kim @sehkone @minshao @msk
Give your opinion on this issue.

[msk]

I'm okay with points 1, 2, and 4 in @syncpark's proposal for the model storage. Point 3, however, is outside the scope of this project. I suggest that this aspect be discussed in the review-web project.

[minshao]

1. On condition that the model could get large, it is indeed a better solution to store them as file and store reference in the database. Further, if the current usage of pretrained collides, then we can store model files there, otherwise, it is better to have a user configurable path designated for storing model file.
2. and 4. both look good to me.

[MW-Kim]

As the analysis continues to reconverge, the actual model size keeps increasing. It would be a good idea to store and manage models in the pretrained directory rather than in PostgreSQL.
The following is the model size for a site with high traffic.

postgres=# SELECT pg_size_pretty(pg_table_size('model'));
 pg_size_pretty
----------------
 1193 MB
(1 row)