Currently we can use conditions: [failure] but not conditions: [error].

The initial driver for this is when using concourse to redeploy itself - see #51. Currently the redeploy step will error, although the underlying bosh deploy succeeds. This feature would enable an automatic retry, which would be a no-op and hence the overall job succeeds.

However, there's a more general use case for this feature - it could be useful to run specific tasks if/when others error (e.g. notifications, or cleanup)