Add 100k read benchmark & client/server bench variants

[alexheretic]

Expand the write bench to have server & client variants and include a mix of binary & text messages to more broadly reflect overall performance.

Add a new "read" bench to similarly bench read performance of client & server. There is no simulated delay in the read "Stream" implementation so the latency should all be from the crate.

Example output: cargo bench --bench \* -- --quick --quiet --noplot

read+unmask 100k small messages (server)
                        time:   [13.611 ms 13.757 ms 13.793 ms]

read 100k small messages (client)
                        time:   [10.041 ms 10.295 ms 10.359 ms]

write 100k small messages then flush (server)
                        time:   [3.6911 ms 3.7038 ms 3.7070 ms]

write+mask 100k small messages then flush (client)
                        time:   [5.2245 ms 5.2252 ms 5.2283 ms]