skip prometheus-client 0.10.0

we saw exceptions on prod that I think might have caused a worker that
is being terminated to die ungracefully. While I'm not sure if this is
an actual problem that changed behaviour (app instances crashing and
restarting), at the very least it definitely polluted the logs and
obscured any actual issues we were having.

https://github.com/prometheus/client_python/releases/tag/v0.10.0

see the pending fix for this problem here:
prometheus/client_python#644

requirements.in

@@ -28,5 +28,6 @@ git+https://github.com/alphagov/notifications-utils.git@44.1.0#egg=notifications
 git+https://github.com/alphagov/govuk-frontend-jinja.git@v0.5.8-alpha#egg=govuk-frontend-jinja==0.5.8-alpha
 
 # gds-metrics requires prometheseus 0.2.0, override that requirement as later versions bring significant performance gains
-prometheus-client==0.10.0
+# version 0.10.0 introduced exceptions when workers crashed due to deprecating lower case `prometheus_multiproc_dir`.
+prometheus-client>=0.9.0,!=0.10.0
 gds-metrics==0.2.4

requirements.txt

@@ -118,7 +118,7 @@ packaging==20.9
     # via bleach
 phonenumbers==8.12.20
     # via notifications-utils
-prometheus-client==0.10.0
+prometheus-client==0.9.0
     # via
     #   -r requirements.in
     #   gds-metrics