Fix haproxy non-agent healthchecks to test something that requires both the

postgrest-rewriter and the postgrest instance to be working.  Our previous
checks just tested whether the rewriter was working, and it could still
succeed even when the postgrest instance it was proxying was down.