Honor timeout in HTTPClient

[d-a-v]

fix #6033
(bug is reproduced and fixed)

[earlephilhower]

I don't see where the write timeout is going to be different between the versions. The online diff seems to have lost its mind, but manually eyeballing I'm not seeing any new checks. What did I miss?

[d-a-v]

What did I miss?

It's not that tricky. Timeout is implemented by Stream::readBytes() and only there.
The issue is the same in HTTPClient::writeToStreamDataBlock() and the fixed example.

readByte()'s timeout was never triggered because of the use of available() which ensured that all requested data would be readable.
The PR avoid the use of available() and ask for what's really requested. The timeout mechanism then can work.

Check the example, and this more readable subcommit.

edit: This was tested with the (now-)fixed example and a web server using netcat offering an incomplete answer in an unclosed connection.

[earlephilhower]

Looks good to me. Only minor niggle is that Stream::readBytes can return 0 for either timeout or connection loss. The handling in these cases looks good, and I suppose a dropped connection would qualify as a timeout (since you'll never get any more data). :)

[devyte]

Consider using above:

std::unique_ptr<uint8_t[]> buff(new uint8_t[buff_size]);

above instead of the malloc, and removing the three free(buff) calls