Keep active ciphers at the front and add metrics

[fortuna]

With this change the active ciphers will be kept at the front of the cipher list. This way the search latency is proportional to the number of active ciphers, rather than total number of ciphers. In practice only a fraction of ciphers in a server is active at any given time.

With this change I no longer shuffle the ciphers.

Benchmark is about the same:

BenchmarkTCPFindCipher-12    	    1000	   1334343 ns/op	  126177 B/op	    1706 allocs/op
BenchmarkUDPUnpack-12        	    2000	    909550 ns/op	  125926 B/op	    1702 allocs/op

The time to find the cipher can be tracked with shadowsocks_time_to_cipher_ms. Example:

# HELP shadowsocks_time_to_cipher_ms Time needed to find the cipher
# TYPE shadowsocks_time_to_cipher_ms summary
shadowsocks_time_to_cipher_ms{access_key="user-0",location="",proto="tcp",status="OK",quantile="0.5"} 0.136696
shadowsocks_time_to_cipher_ms{access_key="user-0",location="",proto="tcp",status="OK",quantile="0.9"} 0.226364
shadowsocks_time_to_cipher_ms{access_key="user-0",location="",proto="tcp",status="OK",quantile="0.99"} 0.226364
shadowsocks_time_to_cipher_ms_sum{access_key="user-0",location="",proto="tcp",status="OK"} 0.6387840000000001
shadowsocks_time_to_cipher_ms_count{access_key="user-0",location="",proto="tcp",status="OK"} 4
shadowsocks_time_to_cipher_ms{access_key="user-1",location="",proto="tcp",status="OK",quantile="0.5"} 0.138071
shadowsocks_time_to_cipher_ms{access_key="user-1",location="",proto="tcp",status="OK",quantile="0.9"} 0.225075
shadowsocks_time_to_cipher_ms{access_key="user-1",location="",proto="tcp",status="OK",quantile="0.99"} 0.302819
shadowsocks_time_to_cipher_ms_sum{access_key="user-1",location="",proto="tcp",status="OK"} 1.6076639999999998
shadowsocks_time_to_cipher_ms_count{access_key="user-1",location="",proto="tcp",status="OK"} 11