Initialize openstack_api_request_errors_total counter to zero (0)

[MPV]

This could help writing queries like:
rate(kos_openstack_api_request_errors_total[5m]) / on (request) rate(kos_openstack_api_requests_total[5m])

...which could be useful for determining an error rate "per request" (and not just an overall error rate, which could be skewed by varying availability in different underlying OpenStack APIs).