Added new image for anomaly plot in matplotlib > 3.4

[jamesp]

This adds a new version of the anomaly plot, added to imagerepo.json in SciTools/iris#4108.

I hope I have done this correctly, please let me know if not

[rcomer]

I think that linscale should have been set to 0.1 rather than 1. This way the scale looks more sensible to me, and closer to what is in our stable docs.

This setting doesn't get the test passing though, so we would need new hashes regardless!

[trexfeathers]

To confirm my understanding:

We had another recent change to this image test in #40. This claimed to be for Matplotlib v3.4.1, but the CI has actually been running against Matplotlib v3.3.4 until more recently. Now that CI is running against v3.4.1, a further minor behaviour change can be seen, necessitating a further change to the test target.

On a different thread: I would prefer to see @rcomer's suggestion a reality, as this would indeed get things back to how they were before #40. I see now that I should not have merged #40; with no prior knowledge, I had assumed @bjlittle's change there was the closest we could get in the new world, but clearly I was wrong.

[rcomer]

In fact, if I set linscale to 0.01, the gallery test passed 👍 . Got a bit confused about exactly what linscale does. I think the key thing is that the value is arbitrarily small.

[jamesp]

Closing this issue as it's not the right fix! Let's open a separate PR on iris with @rcomer's suggested change above