Improve performances of Bloop/install

[Baccata]

Prevents having to iteratively querying the location of each artifact's resolved files. This makes a big difference in large mill projects.

[lolgab]

Does this close #4469 ?
If not it seems for sure related.

[lihaoyi]

CC @alexarchambault and @lefou , not sure if you have any opinions on this since you know more about IDE stuff than I do. Otherwise it looks good to me and hopefully the existing tests catch any bugs

[Baccata]

Does this close #4469 ?

I'm honestly not sure, but I've tried this change on a large mill project that was certainly experiencing similar symptoms, and it made it better.

[lefou]

Looks reasonable to me.

[alexarchambault]

This switches that bit of code from the v1 to the v2 API of coursier, which is nice. The former code was apparently running the resolution twice, so that's better now 👍

[Baccata]

Arguably it could be made event better if the task graph of java/scala modules were keeping the coursier dependencies around alongside the paths to the downloaded artifacts ... but 🤷‍♂️

[alexarchambault]

Arguably it could be made event better if the task graph of java/scala modules were keeping the coursier dependencies around alongside the paths to the downloaded artifacts ... but 🤷‍♂️

#4608 adds helpers to get artifact details without calling too directly the coursier API, that could help here.

We could try adding tasks keeping all those details, but I fear these might be quite heavyweight 🤔