return scale loss

[Ghostvv]

Proposed changes:

• fixes Any intent without an extracted entity fails to be classified correctly after upgrading to 1.8.1 #5451
• add loss scaling to entity loss
• do not scale mask loss
• only scale if some of the examples in a batch has probability of the golden label more than 0.5

Status (please check what you already did):

• added some tests for the functionality
• updated the documentation
• updated the changelog (please check changelog for instructions)
• reformat files using black (please check Readme for instructions)

[tabergma]

👍 Looks good!

[tabergma]

Also, I think we should add a changelog entry.

[Ghostvv]

@tabergma I updated changelog and changed the logic of scaling loss: now it only turns on, if at least 1 example in a batch is learned (making training loss curve jump to higher values after some epoch)

don't know what to do with DeepSource check: I didn't change anything regarding the bugs, it raises

[tabergma]

DeepSource checks the complete file you touched. If the checks are not critical, I guess you can ignore them for now. I think we are fine here. But ideally all of them are tackled.