[M1] Performance optimizations (if necessary)

[matt-codecov]

Ideas that come up during M1 that we want to throw somewhere so they aren't forgotten.

Matt's ideas

• Throw inserts on a SQLite worker thread
• Get rid of joint PK in measurement tables in favor of rowid [wip] get rid of joint PK, just use local_*_id codecov-rs#15
• Make chunks parser a lazy iterator over LineSessionModelSet (or whatever the type is) (example)
  • Gets the side effects out of the parser
  • Batches inserts more efficiently; you might be able to get through 5 or 10 chunks before needing to insert and start a new batch
  • May mean not using winnow anymore, or using it less

Arpad's ideas

• Use serde more Parse report_json files with serde codecov-rs#18