Constant Database native golang implementation

Hi, @alldroll !

I removed few allocations in this PR. And added KeyBytes, ValueBytes methods to public api of Record. These methods don't requires extra allocation for io.NewSectionReader and they are simple to use.

Here is benchmark result:

BenchmarkGetReader-8              581259              2252 ns/op            4352 B/op          2 allocs/op
BenchmarkReaderGet-8              693465              1508 ns/op              64 B/op          6 allocs/op
BenchmarkWriterPut-8              312033              3523 ns/op             283 B/op         16 allocs/op
BenchmarkIteratorAt-8             777184              1369 ns/op             185 B/op          9 allocs/op
BenchmarkIteratorNext-8          2015468               593 ns/op               8 B/op          0 allocs/op
BenchmarkIteratorRecordKey-8     1315321              1072 ns/op             129 B/op          5 allocs/op
BenchmarkIteratorKey-8           2032114               553 ns/op               1 B/op          1 allocs/op

Old benchmark result:

BenchmarkGetReader-8              334984              3207 ns/op            4352 B/op          2 allocs/op
BenchmarkReaderGet-8              440910              2632 ns/op              64 B/op          6 allocs/op
BenchmarkWriterPut-8              158534              8599 ns/op             279 B/op         16 allocs/op
BenchmarkIteratorAt-8             404857              2549 ns/op             185 B/op          9 allocs/op
BenchmarkIteratorNext-8          1831125               771 ns/op               8 B/op          1 allocs/op

## Release notes:

### Added
* `Key()` and `Value()` methods in `Iterator` interface. It is usually easier to use and faster then `iterator.Record().Key()`. Because it doesn't requiers allocation for record copy. But `iterator.Record().Key()` is lazy. It means, a data is read on require.
```
BenchmarkIteratorRecordKey-8     1315321              1072 ns/op             129 B/op          5 allocs/op
BenchmarkIteratorKey-8           2032114               553 ns/op               1 B/op          1 allocs/op
```

### Changed:
* `cdbReader.Iterator()` method returns `ErrEmptyCDB` for empty cdb files
* `cdbReader.Get(key)` method returns `ErrEntryNotFound` if `key` was not found in cdb file

I've refactored tests with testify. Seems it looks nice now.

Unfortunately, I cant do anything with benchmarks. There is no support of benchmarks in testify/suite.

Hi!

CDB file mmapping could improve performance. I wouldn't recomend to use it with fat CDBs.

I benched it here and got 7 times performance improvement on my laptop.

BenchmarkReaderGet-8          	  759301	      1397 ns/op
BenchmarkReaderGetMMaped-8    	 6424041	       194 ns/op

May be we should add this point to README.md?

mmapping is better than using of tmpfs because mmapped file reading does not requre any syscall

#2 I fix to be using io.Copy from reader stream to byte.Buffer instead of making slice object. And I would like to indicate memory usage performance. Let me how to measure the performance which you tested.

You have moved github.com/alldroll/go-datastructures/cdb

https://github.com/alldroll/go-datastructures/commit/0f1c7859cbe52838a8a2faf7b69b82d51d1915bf

Record Key, Value methods return io.Reader with corresponding size. Every loop step allocates memory to read the key and value. Reusing already allocated memory should help with this issue.