The real problem here is the sidecar is picking up:
image: 'prom/haproxy-exporter:latest'

That is 'latest', not the specific version that runs with your version of OCP. When 'prom/haproxy-exporter' was updated, every existing router with the sidecar will fail on restart. It also means the 'latest' sidecar will be running on older OCP versions after restart.

The fix for existing routers is to edit the router DCs that are running the side car with the version that runs with the installed OCP version.

From: https://github.com/openshift/origin/issues/15982

prom/haproxy-exporter:v0.7.1 is for the -haproxy.scrape-uri

and

prom/haproxy-exporter:v0.8.0 is for the --haproxy.scrape-uri

This fix will allow 'oc adm router' to create new routers that will use the 'latest' sidecar.

The sidecar is deprecated and replaced by the integrated prometheus scraper, so it will be going away in a future release.

Going to backport 
https://github.com/openshift/origin/issues/15982
to 3.6.1

Actually, here is the change:

https://github.com/openshift/origin/commit/8573e88f77207f38abd59b70ab0c4fff9a61d632
Handle the changed --haproxy.scrape-uri argument (- to --)

The argument changed from -haproxy.scrape to --haproxy.scrape.  This
makes 'oc adm --expose-metrics' handle the change.

Fixes bug 1488954 (https://bugzilla.redhat.com/show_bug.cgi?id=1488954)

PR https://github.com/openshift/ose/pull/888

Verified this bug on openshift v3.6.173.0.59

this issue has been fixed.

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHSA-2017:3389