Flush to disk faster beyond some number of finalized blocks

[Scooletz]

When processing a lot of blocks that are already finalized and beyond a given threshold M some blocks could be squashed together or omitted in other way so that they do not incur that much IO.

Potential solutions:

1. change the flusher in Blockchain so that if it notices that there's more than N blocks (N - history depth) it squashes N blocks applying them in the same transaction. This would reduce the number of writes greatly without changing the way blocks are flushed.
2. make the flusher use approach like Paprika.Importer where the memory mapped file is not written at all only to be flushed at the end
3. others...

[Scooletz]

I think that with 1st, we can go safe and fast. Especially if the marking as finalized happens in large chunks. Squash every kN + 1 (+1 to move to another slot) and write only then

[Scooletz]

This should not be the issue right now because of two things from general pov of operations.

1. Batching file writes #296 that reduces a lot of syscalls, meaning, the writes should be much faster

2. Paprika/src/Paprika/Chain/Blockchain.cs

   Line 152 in 47a32fb

   var level = reader.TryPeek(out _) ? CommitOptions.DangerNoFlush : CommitOptions.FlushDataOnly;

   that skips the FSync as it's most time consuming where the queue has a lot of these.