Fix regression from 6379b473

[reynir]

The server-sig-algs should be Hostkey.preferred_algs, not just the algorithm for the servers host key. That way we can have a server that uses a ed25519 host key and allow authentication with an RSA key.

Found while testing #74 using awa_test_server.

[hannesm]

this is a partial revert of 6379b47 - I believe your code is correct, but I wonder whether the other part (in let rekey...) should be reverted as well? I don't know exactly how I ended up with 6379b47...

[reynir]

I commented in the commit also. The other usages are for server_host_key_algs in kexinit. It's used there for a different purpose: what algorithms are available for the server host key(s) (of which we currently only have one). So in that case it's desirable to only send algorithms which match our server host key.

[reynir]

On the other hand it seems we had an error in rekeying before that commit as we allow (almost) any server host key sig algs when rekeying?! I don't remember how rekeying works, but that seems odd if not wrong.

[reynir]

Reading rfc4253 pp. 14, 22-23 it seems an error. It's possible to change the host keys and the algorithms in a rekey.

[reynir]

It seems I am to blame for suggesting this regression :D #62 (comment)