Bench: optimise binary data generation

[alexheretic]

I noticed slice::to_vec seems to be faster than .into() 🤷. This code is also more compatible with the proposed approach in #462

As this generates vec data faster it makes the bench faster (but just the bench) .

write 100k small messages then flush (server)
                        time:   [4.0320 ms 4.0390 ms 4.0672 ms]
                        change: [-8.8302% -8.3994% -7.9684%] (p = 0.05 > 0.05)

write+mask 100k small messages then flush (client)
                        time:   [5.3872 ms 5.3921 ms 5.4115 ms]
                        change: [-8.0997% -7.6006% -7.0983%] (p = 0.07 > 0.05)

[daniel-abramov]

I noticed slice::to_vec seems to be faster than .into() 🤷

Huh! That's very interesting. Would not expect that either!