fixed memory leak of ssl session reuse in dyups and session sticky module

[chobits]

try to fix #1016

• todo: added test case for session reuse into dyups.t
• todo: added test case for session reuse into session_sticky.t

[chobits]

try patch: https://patch-diff.githubusercontent.com/raw/alibaba/tengine/pull/1708.patch firstly

[chobits]

ngx_http_dyups_save_peer_session
   -> ctx->ssl_session = ssl_session;
// ctx is alloced from r->pool, which get reference of ssl_sessoin. 
// After r is released, ssl_session is not freed .
//
// There is another problem, since the ssl_session exists on the current ctx of r.
// After new request r' comes in, there is no ssl session saved on the new r'/ctx'.
// In such case,  ssl session reuse does not take effect

1. Session leaks
2. Session reuse logic does not take effect.

[chobits]

We make sure that (sspd == &sspd->rr). See:

typedef struct {
    /* the round robin data must be first */
    ngx_http_upstream_rr_peer_data_t    rrp;   <- &sspd->rrp
...
ngx_http_upstream_ss_peer_data_t;              <- sspd

(refer to nginx core module: ngx_http_upstream_ip_hash_module.c)

[chobits]

Dont need to re-write set_session() and save_session() function when u->peer.data is pointer of ngx_http_upstream_rr_peer_data_t, see ngx_http_upstream_ip_hash_module.c.

Only when u->peer.data is changed to module private data, it needs to re-write these two callbacks. For example, see ngx_http_upstream_keepalive_module.c or dyups module.

[nandsky]

这个module里有不少code style问题，本次pr要不一起调整了？

[nandsky]

LGTM

[chobits]

这个module里有不少code style问题，本次pr要不一起调整了？

has been merged. May next time