Enable StrictData wherever our tests don't fail

[balacij]

Contributes to #3710, where non-strict data is a very common warning noted by stan.
Contributes to #2873, where we need StrictData to build maps of chunks with 'sane checks on insertion'.

I also manually set the language specification to Haskell2010. That's the default that hpack uses, but I think that being explicit is good here. Aside: I'm also okay with upgrading this to GHC2021 or the more recent 2024 spec.

[balacij]

We were convinced that strict data was the right way to go, right? In light of #4006, there might be a larger discussion to be had.

[JacquesCarette]

Just one oddity. I'm kind of surprised this just works, given that we know there are cycles!

[balacij]

Yep, all the cycles involve data types defined in drasil-(lang|database|code).

One involving drasil-code in particular has me dumbfounded -- it causes a hard type error to be thrown from GOOL, but I haven't found the unused problematic expression yet.

[JacquesCarette]

What puzzles me is that this passes the tests?

[balacij]

Yes, all usage of data types from these packages can be strictly used, it seems. The real problematic area is drasil-database which would force evaluating all thunks in our chunks. That would be a mess. These ones are all guarded by laziness that handle the reference loops nicely.