Improve bytes and bytearray serialization

[jakirkham]

Ensure bytes and bytearray serialization are handled correctly for each type respectively. Also adds a fast path for the common case where only a single frame of the right type is provided. This will also nicely build off of the work in PR ( #4004 ) to improve serialization further. This results in more efficient serialization for these types as result. For example take this case of bytearray serialization before and after this change.

Before:

In [1]: from distributed.protocol import serialize, deserialize

In [2]: b = 1_000_000 * bytearray(b"abc")

In [3]: %timeit deserialize(*serialize(b))
137 µs ± 1.5 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

After:

In [1]: from distributed.protocol import serialize, deserialize

In [2]: b = 1_000_000 * bytearray(b"abc")

In [3]: %timeit deserialize(*serialize(b))
6.37 µs ± 51 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)