always store CH in ConnectionState

Fixes a bug. On a server side, ConnectionState.ClientHello was filled
only in case server was handling TLSv1.3

* removes useless variable

* [sike] Adds X25519-SIKEp503 with cSHAKE

Previously there where two methods used for key agreemnt
tls.Conn::generateKeyShare and tls.Conn::deriveDHESecret. Both were
used on client and server side. Boolean flag is used in order to
differentiate between key agreement performed on client and on server
side. Which sucks badly.
In order to implement shared secret agreement with KEM it is better to
add method which implements server specific key agreement and provide
default implementation which reuses tls.Conn::generateKeyShare followed
by tls.Conn::deriveDHESecret.
Now, it is possible for most of the DH-style key agreements to reuse
default implementation and for KEM-style key agreement to provide server
specific implementation.

Tris tries to connect to BoringSSL over TLS 1.2 with X25519-SIDH
as prefered DH group. As this is not supported by BoringSSL it must
fall back to P-256 (second preference on the list)

Also refactors tris test client

Users are sometimes confused why session tickets are not enabled even if
SessionTicketsDisabled is false.

Change-Id: I3b783d2cf3eed693a3ad6acb40a8003db7e0b648
Reviewed-on: https://go-review.googlesource.com/117255
Reviewed-by: Adam Langley <agl@golang.org>

I was confused about how to start an HTTP server if the server
cert/key are in memory, not on disk. I thought it would be good to
show an example of how to use these two functions to accomplish that.

example-cert.pem and example-key.pem were generated using
crypto/tls/generate_cert.go.

parsePrivateKey can't return useful error messages because it does trial
decoding of multiple formats.  Try ParseCertificate first in case it
offers a useful error message.

Fixes #23591

Change-Id: I380490a5850bee593a7d2f584a27b2a14153d768
Reviewed-on: https://go-review.googlesource.com/90435
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Adam Langley <agl@golang.org>

I don't expect these to hit often, but we should still alert users if
we fail to write the correct data to the file, or fail to close it.

Change-Id: I33774e94108f7f18ed655ade8cca229b1993d4d2
Reviewed-on: https://go-review.googlesource.com/91456
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>

Minimal amount of changes needed to switch to GO 1.11
* crypto/internal/cipherhw was removed and internal/cpu was introduced
* wrong code formatting in handhsake_server_test.go was breaking
  the build

Implements two tests for SIDH/P503-X25519 interoperability. BoringSSL
initiates connection to TRIS and TRIS initiates connection to BoringSSL.
SIDH server always listens on port 7443

* SIDH/P503-X25519
* adds interop tests

generateKeyShare has nothing to do with Config receiver. It doesn't
configure anything. It has everything with connection so moved to Conn.
Making deriveECDHESecret also a method of Conn will make it possible to
access 'isClient' field.

serverHandshakeState::readClientHello was setting selected ALPN protocol
always on hs.hello.alpnProtocol, which is specific to TLS 1.2 and older.
Because of that server was marshalling ALPN to SH instead of EE.

see RFC7627

Before AEAD encryption data are stored in a buffer. Last possition
of this buffer store data type. When subsequent TLS record is shorter
than previous, the buffer is shrinked. This causes to remove data
type, which results in wrong MAC calculation. Only in case of TLS 1.3.

0 is used as a sentinel value when determining if the key share should
be marshalled.

The marshaller correctly checks the key_exchange length, but the test
code sometimes generated empty key_exchanges.