Improve input validation to support Unicode/utf-8 charsets

[toholdaquill]

Currently, input validation only supports ascii chars, e.g.:

def clean(s, also=''):
    # safe characters for every possible word in the wordlist includes capital
    # letters because codename hashes are base32-encoded with capital letters
    ok = ' !#%$&)(+*-1032547698;:=?@acbedgfihkjmlonqpsrutwvyxzABCDEFGHIJKLMNOPQRSTUVWXYZ'
    for c in s:
        if c not in ok and c not in also:
             raise CryptoException("invalid input: %s" % s)
    # scrypt.hash requires input of type str. Since the wordlist is all ASCII
    # characters, this conversion is not problematic
    return str(s)

As part of developing i18n support for SecureDrop, the crypto_util.py clean function should be updated to sanitize Unicode/utf-8 input.

This is important for two reasons:

• if we support codenames in Unicode/utf-8 charsets (issue Support Diceware wordlists in multiple languages as part of i18n efforts #999), then input validation should match the language in question
• If SecureDrop decides to accept Transifex translations (see discussion in comments to issue SecureDrop i18n #753, esp. @runasand's remarks), then those translations could also be validated in the same way.

[robbintt]

The clean function can be removed. This will solve internationalization for word lists.

clean is only called in crypto_util.py -- it is called twice.

scrypt.hash in Python 2 needs str() type. This is easy: http://stackoverflow.com/a/1207836

The clean function does manage conversion from unicode to string, so this must be added elsewhere. Not sure what happens with the error handling for CryptoError now, but this raise should minimally be preserved or removed.

###Security
There is a hanging security issue with unicode homoglyphs: https://en.wikipedia.org/wiki/Homoglyph#Unicode_homoglyphs

This is not a huge deal because:
As long as each word list is just one language there shouldn't be any local homoglyphs.

[eloquence]

This is still a valid issue, and we should resolve in order to address #999.

[nabla-c0d3]

The CryptoUtil.clean() function has been removed as part of #5600.

[eloquence]

Thanks @nabla-c0d3! It looks like we're still doing character validation against a limited character set in a different function now.

https://github.com/freedomofpress/securedrop/blob/develop/securedrop/crypto_util.py#L41-L42
https://github.com/freedomofpress/securedrop/blob/develop/securedrop/crypto_util.py#L318

Therefore keeping this issue open.