Improve Error Messages / Logs

[Timoeller]

We can improve Error messages to be more informative. This is a non comprehensive list, feel free to add incomprehensible error messages in the issue thread:

Prio 1

• Only show for real download: "Could not find deepset/roberta-base-squad2 locally. Try to download from model hub ..."
• Remove: "WARNING - farm.modeling.language_model - Could not automatically detect from language model name what language]it is. ..."
• Error if no labels are produced by processor at training time (probably together with a ascii overview of connections between PH and tasks)
• when passing in an empty list as label list it breaks. passing anything (not matching the labels or even the number of different labels present in data) makes it work. This might brake things later on when evaluating (Run NER)
• passing the wrong train filename produces a cryptic error message: ValueError: Unexpected character found when decoding 'true' (Try QA)
• for QA: do not display all cases where answer text is not euqals answer position. Print out stats how many examples could not be converted and maybe the ids
• deprecate message for inference_from_objects (when we call inference from haystack) (don't call inference_from_objects in haystack)
• Move unimportant messages from info to debug level
• Reduce logged samples from 2 per dataloader (train,dev,test) to 1x for train and 1x for dev

Prio 2:

• "Could not convert labels to ids via label_list!
• If your are running in inference mode: Don't worry!" is displayed for every sample in NER inference mode. Lets find a switch to display it only once.