Fixes issues with unexpected characters breaking things when setting the
cookie (like semicolons).
This change as-is does not affect the security of code-server
itself (we've just replaced the static password with a static hash) but
if we were to add a salt in the future it would let us invalidate keys
by rehashing with a new salt which could be handy.
If you visit /login/ instead of /login the cookie will be set at /login
instead of / which means the cookie can't be read at the root. It will
redirect to the login page which *can* read the cookie at /login and
redirect back resulting in an infinite loop.
The previous solution relied on setting the cookie at / (any invalid
value works) which then overrode the login page cookie since
parseCookies only kept a single value. So the login page would see the
same cookie the root was seeing and not redirect back. However, that
behavior depends on the cookies being in the right order which I'm not
sure is guaranteed.
This new method tests all available cookies and always sets the cookie
so the root path will be able to read it in case the login page is
seeing a cookie the root can't.
It also goes a step further and explicitly sets the path on the cookie
which fixes the case where there is a permanent misconfiguration
redirecting /login to /login/. Otherwise the cookie would continually be
set on /login only and you'd have another loop. It also means you only
need to delete one cookie to log out.
Lastly add some properties to make the cookies a bit more secure.
It doesn't show in the explorer anymore so there's no point. Also remove
the local scheme transform which is no longer required with the latest
client-side extension implementation.
Also too the opportunity to rewrite the build script since there was a
change in the build steps (mainly how the product JSON is inserted) and
to get the build changes out of the patch. It also no longer relies on
external caching (we'll want to do this within CI instead).
