luci-base: cbi.js: fix sfh() hash calculation over multibyte sequences

The C side implementation of the hash operates on bytes while the JS variant
operated on UTF-16 codepoints, leading to miscalculations on input strings
containing multibyte sequences.

Recode the given string to an internal UTF-8 byte representation and
calculate the hash over that.

Fixes client side mapping of translation strings containing non-ASCII
characters.

Signed-off-by: Jo-Philipp Wich <jo@mein.io>