Geo: Avoid having to add every replicator class to a list

Gitlab::Geo::ReplicableModel.replicators requires all Models which include ReplicableModel to be loaded before being called !35443 (comment 376048633), or else Replicator classes may be missing.

The development environment has config.eager_load = false by default, unless you enable ENV['RAILS_PROFILE'].

A Slack convo about this

https://gitlab.slack.com/archives/C7U95P909/p1594316455078000

mkozono: Is this a good idea? I would really appreciate a review so we can progress on !35443 (merged) !36467 (closed)

gabriel: I normally don't like to rely on disk traversal... what problem are we trying to fix?

mkozono: The existing ReplicableModel.replicators method doesn't work if all the models are not loaded, which I guess can happen in dev/test

mkozono: !35443 (comment 376048633)

mkozono: Thanks for the MR comment. I think you're right that this approach is better in an initializer, but I'm not sure it's worth the complexity at that point. I rather vote for hard coding a list of Replicator classes. Thoughts? (Also @valery)

gabriel: we can put the files in autoloader, and that will fix the case where they are not loaded

valery: Yeah, seems like a boring solution

mkozono: Isn't that already happening, but config.eager_load is false in development?

valery: I believe Gabriel meant some (other) preload mechanism

gabriel: no I meant that

gabriel: Last time I checked that we were not default to false in development yet, but I believe even for development there is a work-around/exception list right?

valery: I believe creating exception is easy but I'm not sure we want it because it will require to update it each time we create a new replicator

valery: as we essentially need to load models, not replicators itself

mkozono: in dev, you can set ENV['RAILS_PROFILE'] to true to make eager_load true

gabriel: the list of things we will autoload is basically the "pattern" implemented on the code

gabriel: it's more or less the approach of doing something as an initializer

mkozono: ok, I think since it seems to me that my MR is not truly blocked by this, I'll open a follow up issue to "Resolve sometimes missing Replicator classes in development" and copy this discussion in. WDYT?