Description of problem:
When VM migration failed, migrate host pending_vmem_size and pending_vcpus_count is not cleared although VM does not migrate to this host.

Version-Release number of selected component (if applicable):
Engine: rhevm-3.5.0-0.17.beta.el6ev.noarch
Host: libvirt-0.10.2-46.el6.x86_64
      vdsm-4.16.7.1-1.el6ev.x86_64
      sanlock-2.8-1.el6.x86_64
      qemu-kvm-rhev-0.12.1.2-2.448.el6.x86_64

How reproducible:
Constantly

Steps to Reproduce:
1. Add 2 hosts and 1 VMs to engine.
2. Migrate VM from host 1 to host 2.
3. During migration progress, switch host 2 to maintenance mode (In order to 
   simulate migration failure).
4. From engine DB, Observe engine pending_vmem_size and pending_vcpus_count of  
   host 2:
   # su postgres
   bash-4.1$ psql engine
   psql (8.4.20)
   Type "help" for help.
   engine=# select vds_name,pending_vmem_size,pending_vcpus_count from vds;
5. Activate host 2 and repeat steps 2-4 few times. 

Actual results:
pending_vmem_size and pending_vcpus_count increased on each migration failue, eventually migration cannot be performed due to host insufficent memory.

Expected results:
Engine should clear pending_vmem_size and pending_vcpus_count after migration failure.

Additional info:
Engine log attached (scenario started at 2014-10-26 10:15:25)