replace db_connection with rw lock

[ruslandoga]

This PR begins to explore RW-lock approach suggested in #192 (comment)

Here's what I plan to try next:

• should Exqlite.RWConnection.begin_transation(conn) or Exqlite.RWConnection.transaction(fun) be supported?
• should gen_statem be used?
• should there be a cache for prepared statements?

[josevalim]

should Exqlite.RWConnection.begin_transation(conn) or Exqlite.RWConnection.transaction(fun) be supported?

IMO, no transaction helpers necessary, unless it is required later by EctoSQLite3.

should gen_statem be used?

Up to you :)

should there be a cache for prepared statements?

I would postpone until there is a need, to keep the first PR small.

[warmwaffles]

This module could be deleted and ignored until you are completely done. I don't know of anyone using this module directly.

[warmwaffles]

should there be a cache for prepared statements?

I would postpone until there is a need, to keep the first PR small.

I remember starting to cache the prepared statements, but then bailed on it due to the complexity of it. I figured since this is all in the NIF it was better to just create it on the fly and then let the garbage collector sweep it away once done.

[ruslandoga]

Closing this PR as I think the processes can be moved to ecto_sqlite3 and exqlite would be just a NIF. I'm not 100% sure, but I'm exploring this approach in https://github.com/ruslandoga/sqlite.ex and https://github.com/ruslandoga/ecto_sqlite which a re basically copy-pastes of exqlite/ecto_sqlite3 with some changes here and there (and different project / namespace makes benchmarks easier).