Always disable sync in SQLite cache

[ilevkivskyi]

Fixes #21176

This seems to resolve all the HDD performance issues. Although this makes cache less durable, this is fine, as FS is not perfect in this sense either (and it doesn't really need to be that durable).

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[JukkaL]

I guess we could probably get away with having synchronous enabled on macOS, since they all have had SSDs for a long time now, but since we'll need this for parallel checking, it makes more sense to keep things consistent.

[cburroughs]

this makes cache less durable

This disables fsync, correct? So this is not just "your recent writes might be missing", but "your file might be corrupted". (I'm not saying this is a bad tradeoff for a cache, but it isn't obvious to me from the linked issues how mypy would handle corrupt sqlite files.)

[ilevkivskyi]

This disables fsync, correct?

I don't know the details, but probably yes.

how mypy would handle corrupt sqlite files

It will likely just crash on startup. If this will become a problem we can detect this condition and manually delete the cache.

[JukkaL]

I agree that implementing automatic recovery from a corrupt cache could be worth it. However, I believe disabling fsync mostly impacts situations where power is lost or the system is hard rebooted -- just killing mypy shouldn't be a problem?

[cburroughs]

just killing mypy shouldn't be a problem?

That matches my expectation.