Fix evaluate() printed values being incorrect by default

[Carath]

Temporary fix for most users to this issue which was not present in previous versions of Keras.

This changes to 2 the default verbosity level of the model evaluate() method, as to not print incorrect values for e.g. the loss and accuracy (mismatching correct values printed during training or values computed "by hand").

This does not fix the problem as to why verbose=1 makes evaluate() print incorrect values while still returning correct ones when passing return_dict=True.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.04%. Comparing base (90568da) to head (46355bb).

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #20799      +/-   ##
==========================================
+ Coverage   82.01%   82.04%   +0.03%     
==========================================
  Files         557      557              
  Lines       52016    52016              
  Branches     8037     8037              
==========================================
+ Hits        42659    42675      +16     
+ Misses       7403     7387      -16     
  Partials     1954     1954              

Flag	Coverage Δ
keras	81.86% <ø> (+0.03%)	⬆️
keras-jax	64.25% <ø> (+0.03%)	⬆️
keras-numpy	58.97% <ø> (+0.01%)	⬆️
keras-openvino	29.89% <ø> (ø)
keras-tensorflow	64.82% <ø> (+0.03%)	⬆️
keras-torch	64.20% <ø> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[fchollet]

If you want to investigate the reason for the discrepancy between the final value printed by evaluate() in the progress bar, and the values returned by evaluate() (which are the correct ones), you are welcome to do so. I'm thinking it might be the progress bar callback being delayed by one batch or something like that.

However disabling logging is not the way to do it.