Better example of simple-encoders/decoder

[spikebike]

I was tinkering with the examples and as a bit surprised that a single bit error can corrupt the decoding. The README does mention this "If values have changed in a shard, it cannot be reconstructed".

I was thinking it might be useful to have an example closer to a real world use case.

The example README does mention hashing the shards so you can delete the corrupt ones. But in the default example (4 shards + 2 parity) even with a hash of each shard, 3 single bit errors could kill a decoding.

So I was thinking of adding a sha256 per 64KB block. That way any bit corruptions would be recognized and you could use the other 5 shards for that particular block.

I wrote a little simulator to add errors and I believe that on average a 2GB file encoded with the defaults (4 shards + 2 parity shards) would survive (on average) between 280 and 1200 64KB errors before any part of the file is corrupted. Quite a bit better than 3 bit errors. Storage overhead would be something like 0.05%. Additionally even with 8192 errors 95% of the file would be recoverable.

Would you accept a pull request to add sha256 block checksums to your simple encoder/decoder?

[klauspost]

What do you think is missing in the current ones?

[spikebike]

Original issue edited.

[klauspost]

@spikebike Feel free to open a PR if you still have the better example.