contour: Mirroring doesn't work correctly with weights

I specified this proxy:

apiVersion: projectcontour.io/v1
kind: HTTPProxy
metadata:
  name: artifactory-v2
spec:
  routes:
    services:
    - name: artifactory-prod
      port: 80
      weight: 95
    - service: artifactory-dev
      port: 80
      weight: 4
    - service: artifactory-sampler
      port: 80
      weight: 1
      mirror: true

And my expectation was that 1% of traffic would be mirrored to artifactory-sampler. However, the envoy configuration ends up like this:

         "routes": [
            {
              "match": {
                "prefix": "/v2/token/"
              },
              "route": {
                "weighted_clusters": {
                  "clusters": [
                    {
                      "name": "default/artifactory-dev/80/da39a3ee5e",
                      "weight": 4
                    },
                    {
                      "name": "default/artifactory-prod/80/da39a3ee5e",
                      "weight": 95
                    }
                  ],
                  "total_weight": 99
                },
                "request_mirror_policy": {
                  "cluster": "default/artifactory-sampler/80/da39a3ee5e"
                }
              }
            },
            {
              "match": {
                "prefix": "/v1/token/"
              },
              "route": {
                "weighted_clusters": {
                  "clusters": [
                    {
                      "name": "default/artifactory-dev/80/da39a3ee5e",
                      "weight": 4
                    },
                    {
                      "name": "default/artifactory-prod/80/da39a3ee5e",
                      "weight": 95
                    }
                  ],
                  "total_weight": 99
                },
                "request_mirror_policy": {
                  "cluster": "default/artifactory-sampler/80/da39a3ee5e"
                }
              }
            }
          ]
        },

Which, IIUC, sends mirrors all the traffic from the route. If weights can’t be combined with mirroring, we should emit a status condition to tell the user that. However, I think that we can look at the weights and automatically set a runtime_fraction for the mirror, so that this behaves the way it looks.

https://www.envoyproxy.io/docs/envoy/latest/api-v2/api/v2/route/route.proto#route-routeaction-requestmirrorpolicy