equality for packages: ordered list or set-based?

The [Spork regression job](https://ci.inria.fr/sos/job/Spork/) might have detected a regression.
This relates to equality of packages.

Actual behavior:
the visiting order is taken into account, a package with class A and class B is considered not equal to a package with class B and class A.

Expected behavior:
we do a set-based equality of contained types.

AFAICS, this is due to the usage of BiScanner.

FTR: current workaround in Spork https://github.com/ASSERT-KTH/spork/commit/0b9e85829fe36cf4b951711110ec27de273cf631