JS output filter silently truncates UTF-16 encoded external .js files

[e.fryntov]

Psiphon applies charset transcoding only to (x)html documents. JS documents have to be transcoded the same way in order to apply search&replace regex rules:

server -> get JS document original encoding -> UTF-8 -> search&replace -> back to JS document original encoding -> browser

Also, we need to to extend the encoding detection by checking for BOM in case there is no charset info neither in the response header nor in the document body(there is no meta tags in JS documents for example)