1. Filter with the regex - what's left has a valid format, making step 2 much saner.
2. Extract and validate the domain name - super simple now, because the domain component is known to be sane.
(Optional but good idea 3: Handle exceptions....)
Step 1 is almost always the hardest part, now it's mostly done.
If it is 80% good, then, yeah, your underlying assertion is likely good: Too little coverage to be all that useful. But 99+% coverage? For a quick pass filter? That's pretty good, we can work with those numbers.
After all, one could have multiple layers of progressively more accurate but progressively slower filters. Not an uncommon or unusual approach.
Besides, the JS version could be very useful as a quick pass usability check, validating whether the user made a silly mistake. And adding some logic to record the bogus entry, compare it to what they write next, pass it to the server if need be, can all be very useful.
foo@bar could still receive email, if you had a host in your DNS domain named bar, with a user named foo.
As I have told people for many years: if you must do this, check at most if there's an @ and (perhaps) a dot somewhere after the @, which is enough to stop someone who has accidentally put their name in the email address field, or a similar user error. Anything else is a waste of brainwidth and will result in more problems than it solves.