cat: make it concurrent

[denizsurmeli]

This PR adds a new io.WriterAt adapter for non-seekable writers. It uses an internal linked list to order the incoming chunks. The implementation is independent from the download manager of aws-sdk-go, and because of that currently it can not bound the memory usage. In order to limit the memory usage, we would have had to write a custom manager other than the aws-sdk-go's implementation, which seemed unfeasible.

The new implementation is about %25 percent faster than the older implementation for a 9.4 GB file with partSize=50MB and concurrency=20 parameters, with significantly higher memory usage, on average it uses 0.9 GB of memory and at most 2.1 GB is observed. Obviously, the memory usage and performance is dependent on the partSize-concurrency configuration and the link.

Resolves #245.

[denizsurmeli]

I have reduced the test size in here.

s5cmd/e2e/cat_test.go

Line 24 in 5b55085

contents, expected := getSequentialFileContent(16 * mb)

The reason for this is that since we have more test cases, we should aim to show that the feature/fix works while maintaining a reasonable time for the tests to run.

s5cmd/e2e/cat_test.go

Lines 50 to 75 in 5b55085

	{
	name: "cat remote object with lower part size and higher concurrency",
	cmd: []string{
	"cat",
	"-p",
	"1",
	"-c",
	"10",
	},
	expected: expected,
	},
	{
	name: "cat remote object with json flag lower part size and higher concurrency",
	cmd: []string{
	"--json",
	"cat",
	"-p",
	"1",
	"-c",
	"10",
	}, expected: expected,
	assertOps: []assertOp{
	jsonCheck(true),
	},
	},
	}

Lowering the file size and adjusting the partSize and concurrency on the tests both show the feature works correctly while maintaining a feasible duration for the test.

[igungor]

🥇