Add TCP_NODELAY option

[oraoto]

During benchmark, I found there are significant delay when the block is smaller than the buffer size (defaults to 8192 bytes).

Test program:

#include <clickhouse/client.h>
#include <cstdlib>
#include <chrono>
#include <iostream>

using namespace clickhouse;

int main(int argc, char **argv)
{
    Client client(ClientOptions().SetHost("localhost").SetPassword("qwerty"));

    client.Execute("CREATE TABLE IF NOT EXISTS test.delay (id UInt8) ENGINE = Memory");

    int count = atoi(argv[1]); // count

    for (int i = 0; i < 10; i++) {
	Block block;

	auto id = std::make_shared<ColumnUInt8>();

	for (int j = 0; j < count; j++) {
	    id->Append(65);
	}
	block.AppendColumn("id"  , id);

	auto start = std::chrono::steady_clock::now();
	client.Insert("test.delay", block);
	auto end = std::chrono::steady_clock::now();

	std::cout << std::chrono::duration_cast<std::chrono::microseconds>(end - start).count() << " ms" << std::endl;
    }

}

When data (8171 bytes + some header) is larger than bufffer:

./test-insert-delay 8171
434 ms
567 ms
366 ms
252 ms
354 ms
265 ms
630 ms
237 ms
234 ms
181 ms

When data is smaller than bufffer, there are significant delay:

./test-insert-delay 8170
213 ms
155 ms
43767 ms
43270 ms
46539 ms
46627 ms
43114 ms
43273 ms
394 ms
42893 ms

To resolve this problem, I add an option to enable TCP_NODELAY. It's enabled by default, same as clickhouse-client.

[oraoto]

Well, the time scale should be μs, which means Nagle's algorithm adds about 40 ms delay.