Separate logic from UI, secure logic with tests

[andreasabel]

Prompted by

• fix(server): email address validation #1074

Validators are working directly in the server monad, e.g.:

hackage-server/src/Distribution/Server/Util/Validators.hs

Lines 23 to 30 in 1c35e03

	-- Make sure this roughly corresponds to the frontend validation in user-details-form.html.st
	guardValidLookingEmail :: T.Text -> ServerPartE ()
	guardValidLookingEmail str = either errBadEmail return $ do
	guard (T.length str <= 100) ?! "Sorry, we didn't expect email addresses to be longer than 100 characters."
	guard (T.all isPrint str) ?! "Unexpected character in email address, please use only printable Unicode characters."
	guard hasAtSomewhere ?! "Oops, that doesn't look like an email address."
	guard (T.all (not.isSpace) str) ?! "Oops, no spaces in email addresses please."
	guard (T.all (not.isAngle) str) ?! "Please use just the email address, not \"name\" <[email protected]> style."

This makes it hard to test them cheaply.

There is a purely functional core that struggles to get out here: the actual validation does not need IO etc., it is pure logic.
Suggested restructuring:

• data FooError = FooProblem1 | FooProblem2 | ...
• checkFoo :: Foo -> Either FooError () does the actual check, indicating the violation by a Left FooProblem...
• guardFoo :: Foo -> ServerPartE () calls checkFoo and handles the exception Left ... by escalating the problem to the user.
• Add unit tests for checkFoo