perf(vector): updated marshalling of vector

[harshil-goel]

Earlier we were unmarshalling bytes to []float64 by iterating on each element and reading it little endian. But we are now doing it using unsafe pointers. This reduces thee time from O(size(bytes)) to O(1) basically.
Benchmark stats:

goos: linux
goarch: amd64
pkg: github.com/dgraph-io/dgraph/tok/index
cpu: 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz
BenchmarkEncodeDecodeUint64Matrix/Binary_Encoding/Decoding-8              185269              5991 ns/op
BenchmarkEncodeDecodeUint64Matrix/Gob_Encoding/Decoding-8                  46299             25547 ns/op
BenchmarkEncodeDecodeUint64Matrix/JSON_Encoding/Decoding-8                 77666             13723 ns/op
BenchmarkEncodeDecodeUint64Matrix/PB_Encoding/Decoding-8                  500092              2346 ns/op
BenchmarkEncodeDecodeUint64Matrix/Unsafe_Encoding/Decoding-8          1465863          826.2 ns/op
BenchmarkDotProduct/vek:size=96000-8                                      528499              2331 ns/op
BenchmarkDotProduct/dotProduct:size=96000-8                               170630              7765 ns/op
BenchmarkDotProduct/dotProductT:size=96000-8                              145855              8314 ns/op
BenchmarkFloatConverstion/Current:size=96000-8                          284873474                4.172 ns/op
BenchmarkFloatConverstion/pointerFloat:size=96000-8                     263052618                3.988 ns/op
BenchmarkFloatConverstion/littleEndianFloat:size=96000-8                   40446             35287 ns/op

Now indexing 500k vectors take about 5 minutes. (more than 5 hours before)

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ harshil-goel
❌ Harshil goel

---

Harshil goel seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[mangalaman93]

Does this change the disk format? Why can't we use protobuf here like we do for everything we write to disk?

[harshil-goel]

We have to iterate over the list to make it a protobuff. So to marshal and unmarshal becomes a linear task. Hence taking too much time.

[harshil-goel]

BenchmarkEncodeDecodeUint64Matrix/JSON_Encoding/Decoding-8 91692 12608 ns/op
BenchmarkEncodeDecodeUint64Matrix/PB_Encoding/Decoding-8 467174 2221 ns/op
BenchmarkEncodeDecodeUint64Matrix/Unsafe_Encoding/Decoding-8 1609965 748.9 ns/op