[2.6] controller panics collecting metrics

The agent would panic with a traceback like this:

panic: label name "workload_status" missing in label map

goroutine 183 [running]:
github.com/juju/juju/vendor/github.com/prometheus/client_golang/prometheus.(*GaugeVec).With(0xc0001e8808, 0xc0001e5c50, 0xb1fb37, 0xf)
        /home/tim/go/src/github.com/juju/juju/vendor/github.com/prometheus/client_golang/prometheus/gauge.go:228 +0x79
github.com/juju/juju/core/cache.(*Collector).updateModelMetrics(0xc0001f7770, 0xb1c009, 0xa)
        /home/tim/go/src/github.com/juju/juju/core/cache/metrics.go:321 +0x533
github.com/juju/juju/core/cache.(*Collector).updateMetrics(0xc0001f7770)
        /home/tim/go/src/github.com/juju/juju/core/cache/metrics.go:292 +0x10a
github.com/juju/juju/core/cache.(*Collector).Collect(0xc0001f7770, 0xc000275380)
        /home/tim/go/src/github.com/juju/juju/core/cache/metrics.go:276 +0x1e7