perf: perf yamux stream read buf

[driftluo]

#398 use extend_from_slice will copy all data twice, this pr use Vec<BytesMut> reduces its copy from all data to just one ptr, and Vec's memory allocation and release do not always work, because drain does not reclaim memory, that is, the memory allocation operation here is limited

bench master and this pr:

master branch:

Gnuplot not found, using plotters backend
yamux_beach/            time:   [475.10 µs 544.57 µs 618.04 µs]                         
                        change: [-5.0316% +12.625% +35.408%] (p = 0.20 > 0.05)
                        No change in performance detected.
Found 4 outliers among 100 measurements (4.00%)
  4 (4.00%) high mild

this pr:

Gnuplot not found, using plotters backend
yamux_beach/            time:   [393.64 µs 451.65 µs 514.75 µs]                         
                        change: [-34.776% -23.230% -8.6939%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  4 (4.00%) high mild