When creating queries from package specs, hy_subject_get_best_solution() does not filter the results according to what package architectures are allowed (which was previously set via dnf_sack_set_arch()).

Normally this is not an issue, as packages with an incorrect architecture will later be rejected anyway through checks happening in libsolv, but in Yocto project we found a scenario where it does lead to an incorrect decision.

Specifically:

1. The system architecture is set to "correct-arch".

2. The installation repository contains two packages:
- package "abc.incorrect-arch"
- package "xyz.correct-arch", with "provides: abc"

3. What then happens is that hy_subject_get_best_solution() will pick "abc" package for the installation (because the code that would consider packages with "provides: abc" is not even reached). Then later the "abc" package will be rejected by libsolv, even though the correct solution would've been to pick "xyz".

More info here:
https://bugzilla.yoctoproject.org/show_bug.cgi?id=12025

I have considered various ways to fix this, but I can't decide which would be the optimal one. This is something for libdnf maintainer to ponder, I guess:

1) Add HY_ONLY_INSTALLABLE to _hy_query_flags, use it in hy-query.c init_result() to filter the package set with the help of pool_installable() from libsolv.

2) Add hy_query_filter_installable() funtcion to hy-query.c, which would use the same pool_installable() to filter the package set. Call it from  hy_subject_get_best_solution().

3) Add HY_PKG_INSTALLABLE type of filter, otherwise similar to solution 2.

4) in hy_subject_get_best_solution() get the list of allowed architectures (libsolv or libdnf do not provide a convenience function for it, but it can be done). 
Then issue:
    hy_query_filter_in(query, HY_PKG_ARCH, HY_EQ, allowed_arches);


Thoughts?