Incorrect model serving with --standardize False

[aguschin]

After #588 it appears that models are sometimes served incorrectly:

def hash_text(text, salt="SALT"):
    return hash(text + salt)


from mlem.api import save

save(hash_text, "hash_model", sample_data="sometext")

then

$ mlem serve streamlit -m hash_model --standardize False
⏳️ Loading model from hash_model.mlem
Starting streamlit server...
🖇️  Adding route for /__call__
Checkout openapi docs at <http://0.0.0.0:8080/docs>
...

and I see

the problem: it's the wrong input to be expected from the model. It should be text. If you input something and click on "Submit", it fails.
In FastAPI:

Interestingly, it works with --standardize True:

[mike0sv]

This should be fixed by #595, please check

[aguschin]

yep! thanks!