blockstore: fast path for AllKeysChan using the index

[MichaelMure]

Note: I'm not entirely sure I didn't break some assumption.

close #242

Crude benchmark:

func BenchmarkAllKeysChan(b *testing.B) {
	path := filepath.Join(b.TempDir(), "bench-large-v2.car")
	generateRandomCarV2File(b, path, 10<<20) // 10 MiB
	defer os.Remove(path)

	bs, err := blockstore.OpenReadWrite(path, nil)
	if err != nil {
		b.Fatal(err)
	}

	b.ReportAllocs()
	b.ResetTimer()

	for i := 0; i < b.N; i++ {
		c, err := bs.AllKeysChan(context.Background())
		if err != nil {
			b.Fatal(err)
		}
		for range c {
		}
	}
}

Before:

BenchmarkAllKeysChan-12 772 2022389 ns/op 89049 B/op 1617 allocs/op

After

BenchmarkAllKeysChan-12 8824 195779 ns/op 30865 B/op 642 allocs/op

10.3X faster, allocate 2.5X less memory.

[MichaelMure]

@rvagg I did what you suggested, the performance boost is even more significant:

Before:

BenchmarkAllKeysChan-12 772 2022389 ns/op 89049 B/op 1617 allocs/op

After

BenchmarkAllKeysChan-12 10000 105874 ns/op 144 B/op 2 allocs/op

19.1X faster, allocate 735X less memory.

[MichaelMure]

Removing the BlockstoreUseWholeCIDs check lead to a CI failure though, should I just put it back?

   === RUN   TestBlockstorePutSameHashes
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      readwrite_test.go:242: expected blockstore without UseWholeCIDs to flatten on AllKeysChan
      ```

[MichaelMure]

Removing the BlockstoreUseWholeCIDs check lead to a CI failure though, should I just put it back?

Answering my own question, that doesn't fix that test. Yet the previous code with the multihash was passing that test ‽

I'll need someone more versed in that code to figure out if that test makes sense, and if so what should be done about it.

[MichaelMure]

@rvagg I applied your changes. Final result:

Before:
> BenchmarkAllKeysChan-12		885	   1256865 ns/op	   88911 B/op	    1617 allocs/op
> BenchmarkAllKeysChanUseWholeCIDs-12	1010	   1253543 ns/op	   57861 B/op	     976 allocs/op

After
> BenchmarkAllKeysChan-12		8971	    135906 ns/op	   30864 B/op	     642 allocs/op
> BenchmarkAllKeysChanUseWholeCIDs-12	13904	     86140 ns/op	     144 B/op	       2 allocs/op

BenchmarkAllKeysChan            --- 9.25X faster, allocate 2.9X less memory
BenchmarkAllKeysChanUseWholeCID --- 14.5X faster, allocate 401X less memory.