[12.0-stable] IoBundle.KeepInHost should not forbid assignment

[milan-zededa]

KeepInHost is used to keep unassigned devices visible to the host, i.e., not reserved for future PCI passthrough. However, once a device with KeepInHost=true is selected for assignment to an application, the assignment takes precedence over KeepInHost. Only IsPort=true devices are forbidden from assignment.

After a recent change, everything is kept in the host until assignment, and therefore we may consider removing the KeepInHost flag. See commit: aad213f

But here we only fix a bug introduced in the commit referenced above. Specifically, KeepInHost was misunderstood and used to forbid (or more precisely skip) device assignment. This makes it impossible, for example, to assign WiFi adapters or cellular modems.

(cherry picked from commit 5b9cbd9)

[eriknordmark]

LGTM

[milan-zededa]

I have added a commit increasing the eden version to the latest.
Many test failures have been fixed recently and I think it makes sense to bring this to 12.0-stable to get greener results on backported PRs.

[milan-zededa]

So we do have 4 locally-reproducible test failures on 12.0 with the latest eden:

 # line 48:
 ./eden test tests/eclient -e app_nonat -v debug
 # line 66:
 ./eden test tests/volume -e volumes_test -v debug
 # line 51:
 ./eden test tests/eclient -e nw_switch -v debug
 # line 76:
 ./eden test tests/eclient -e app_local_info -v debug

They are not related to this PR but to changes done in the sequence of application states when purging.
The problem is that the latest eden was updated to reflect these changes, which were merged to master after 12.0 was released.
So we cannot really pull the latest eden in 12.0. But at the same time I think it would make sense to have green test results also on this LTS branch. I'm removing eden-upgrade commit and will think about how to address this...