Use TCP healthcheck for registry service
From HAProxy we use option httpchk GET /debug/health HTTP/1.1\r\nHost:\ registry.gitlab.com
as a health check.
I don't think this is the right approach as it is possible for us to lose an entire cluster if this check fails, which it might for pods that are in transitional states. We saw this behavior in production#3299 (closed)
I suggest we move this backend to a TCP check, like we do for git https