fix(transport): Send message and separator in one write call

[jarrismwhite]

With multiple workers, a second message can get written before the lineSeparator can, resulting in parsers counting on the separator to fail intermittently. It doesn't happen for every message, but it's enough to measure consistently.

Expected:

"{...}\n{...}\n{...}\n"

Observed:

"{...}\n{...}{...}\n\n"

It seems like this was introduced with #447 when data and lineSeparator were separated into two writes.