fix binary classification prediction from v11

[david-behar]

No description provided.

[Marsobad]

Code LGTM

• Bug reproduced on DSS 11
• Fix tested on DSS 10 + 11
  • train in analysis
  • train + score in the flow
  • custom web app

Regarding integration tests, someone from my team is working on setting them up on a DSS 11 instance.

However, I am worried about customers who upgrade to v11 without updating the glm plugin. If they deployed a glm classification, their existing flows will break => they'll get the following error message in the deployed train recipe : [12:10:30] [INFO] [dku.utils] - Exception: Data must be 1-dimensional (same thing if they retrain from the original analysis)

The least we can do is to document this error in the change log and the plugin doc. We need to think about other solutions to prevent flows from breaking.