Send entire payload considering IO#syswrite return size.

[mashhurs]

Issue description

As a client, when sending over 16K data with SSL enabled mode, it is confirmed that some portion of data is not transferred.
Eg. generated 16507 bytes data, IO#syswrite only sent 16367 bytes

payload = 'A' * (16 * 1024 + 123) # 16k + 123 = 16507

What this PR does?

Takes in a count of IO#syswrite return size and iterates till payload is transmitted.

Closes #41 #30 #33

Testing

Added a logic to generate over 16K data and put some printfs to track the send size.

• Config

output {
    stdout {}
    tcp {
        host => "google.com"
        port => 443
        ssl_enable => true
    }
}

• Logs

Sending data: {"@timestamp":"2022-08-16T18:08:55.282249Z","event":{"original":"test"},"host":{"hostname":"localhost"},"@version":"1","message":"test"}
Sent size: 171

Sending data: {"@timestamp":"2022-08-16T18:09:04.404790Z","event":{"original":"generate"},"host":{"hostname":"localhost"},"@version":"1","message":"generate"}
Generating more than 16K payload...
Sent size: 16367
Sent size: 140 # expectation

• Checked the thread no on the messages

warning: thread "[main]|output|tcp|client_socket-1" terminated with exception (report_on_exception is true):

• Debugged logic
  • TCP server to test: gist
  • TCP client to test: gist

[yaauie]

I realized that my first pass was effectively commenting only on lines that were whitespace-changes from the perspective of this PR, and haven't had time to fully grok the effective changes you're proposing. I'll get back to this tomorrow when my brain isn't a puddle of mush :)

[yaauie]

More discovery and feedback, now that my brain is back online :)

[yaauie]

This PR has gone through a lot of iteration and discovery, but to my eyes it looks like it correctly:

• fixes an issue in both client and server where bytes from large payloads or to streams under back-pressure could be truncated
• fixes a regression from Feat: ssl_supported_protocols (TLSv1.3) #47 in which the logger method wasn't accessible to the client, so it would throw an error while attempting to log a closed stream.
• fixes an issue where client threads were not guaranteed to be distinct

I've left a few nitpicks, but it certainly looks on track to make things better 👍🏼

[mashhurs]

Tested after applying review feedbacks.

[yaauie]

One suggestion on the changelog to better describe the problem that we are fixing, but implementation and improvements all LGTM 👍🏼