It seems that currently there is no way to avoid the use of try/except in at least one of the functions.

This is definitely not expected and should be mentioned in the description.

The tests do rounding to check if the returned value is close enough to the expected value, but the anti-cheat check doesn't, which causes false positive.

Lots of existing solutions, including author solution, fail at certain combination of values by division by zero. These edge cases should be tested as fixed tests.

Looks like you are cheating (18 times) without arithmetic operators (and without import).

What else is forbidden?
Should be specified in description.

Ranges of tested values should be clearly specified. For instance, the reference solution does not work for multiply(0, 1) but it works for multiply(1, 0).

There appears to be no enforcement of the rule.