Build YAML by streaming

[RandomInsano]

This PR changes cargo loco seed --dump to stream data to files instead of building a large string before writing to disk. It uses a buffered writer to lessen the access overhead. I will add the same for seeding, but that will require a bit more work and I wanted to get opinions on this first.

One problem is that I couldn't find a reasonable way to access sqlx's streaming for custom queries in SeaORM, so in my case Loco is still allocating 6GB of RAM for 3 million rows. Is it alright for me to access sqlx directly here?

Sorry for the formatting shift here. Let me know if you want me to revert it.

[RandomInsano]

Did some experimenting here. Consider this an RFC because I've fixed my problem and can polish this up to upstream.

Because serde_yaml didn't support streaming inside of async context with serde_yaml::Serializer, I changed the serializing to serde_json. The YAML serializer failed due to an internal `static lifetime, but it's also deprecated. Frankly, I prefer YAML for this context myself.

The format of the JSON is not a true JSON array but just a series of JSON objects side-by-side. I've confirmed jq is happy with this, and it allows us to use serde_json::Deserializer::from_reader() to read data in object by object. I'm not sure if it's safe/possible to put this into a database transaction, but happy to do that if the import should be atomic (say a dump file is corrupted).

I raised a dump() function into the Hooks trait so that the Generics could be used when serializing. What I don't like at the moment is that the implementor of that function needs to do the table filtering, but I don't have a good solution to promote it to run_app_dump()... Some advice there would be help, but I can mull it over if the work I've done here is worthwhile.