Consider moving from persistent + esqueleto to beam

[lfborjas]

They have:

• Actual support for enums, maybe? -- this would make queries on ZodiacSign and Gender much faster, since we'd no longer have to deal with string comparisons!
• Windowing functions
• Actual support for text search types
• INSERT ON CONFLICT
• JSONB and jsonb_array_elements
• citext Also builds on top of postgresqlsimple. Also, unrelated, I found the quasiquoter that I was looking for: https://hackage.haskell.org/package/postgresql-simple-0.6.2/docs/Database-PostgreSQL-Simple-SqlQQ.html

Try it out in a separate module, defining types for Dream and maybe User, and see if we can get our gnarly queries to be extensible -- that's what worries me the most about esqueleto, its paltry support for fancy postgres features and the early need to fall back to raw SQL, which isn't composable nor type safe.

[lfborjas]

Alternatively, what I really care about is:

• Proper enum support. Maybe we can just write the instances, like this guy does for UUIDs: https://bitemyapp.com/blog/uuids-with-persistent-yesod/
• In fact, my biggest motivation is dynamically composing the tricky "stats" queries. I believe these also fall outside of the beam purview (e.g. I don't think it can deal with the fancy ts_stat($$some-query-here$$) situation either.
• I don't like the compromise on keys that beam makes. Maybe it's stockholm syndrome, but I like Entity Key Dream for example: never having to carry around the primary key with the record!

[lfborjas]

After the improvements in #19 and #20, not inclined to do this just yet.