Fixes issue #912: Executes liftF effect twice in a product composition

[Lost-Fly]

Fixes issue #912

When combining a Done fetch with a Blocked fetch in product/productR/map2, the previous implementation would call product(fa, c) or map2(fa, c)(f), causing the already-completed effect fa to be re-run.

The fix uses the already-computed result value instead of re-running the entire fetch:

• product: use c.map((a, _)) instead of product(fa, c)
• productR: use c or c.as(b) instead of productR(fa)(c)
• map2: use c.map(b => f(a, b)) instead of map2(fa, c)(f)

Added unit tests to verify that side effects are executed exactly once.

[Lost-Fly]

@AntonioMateoGomez @fedefernandez Could someone from the maintainers please take a look when you have a moment?

[sloshy]

Spent some time trying to see if this was representative of a behavior regression by running your tests on older releases from 5+ years ago, but no, I believe that this behavior was just never tested for before. Good find, and I'll double-check with some other Xebia people to see if they concur but I believe this can and should be merged.

[sloshy]

@Lost-Fly if you can run sbt scalafmt for me on this branch I'll get it merged ASAP and hopefully cut a release soon!

[Lost-Fly]

@sloshy I`ve done scalafmt on tests, check it please

[sloshy]

Currently waiting on #915 and then this can be updated with main, and should be good to go after that.

[sloshy]

@Lost-Fly seems like it'll work again once you update the branch from main. I'd do it but then it'd associate me with the commit, and I believe that would mean I'd need someone else to review it.

[Lost-Fly]

I've done rebase onto main. Please run the tests now