Implement fast in-memory database with JSON storage

[faustbrian]

Things like the transaction pool could greatly benefit from a fast in-memory storage that could fallback to store everything in a JSON file for which we then could use fast.js to do very fast queries on objects rather then a database as the pool itself doesn't contain any relational data.

Something like https://github.com/typicode/lowdb but a faster implementation because lowdb uses lodash which is rather fat and slow.

[vasild]

Currently we store the transactions in the transaction pool in memory in a quite efficient way. It is more involved than the following, but to illustrate, we use:

byId = {}
byId[tx1id] = tx1
byId[tx2id] = tx2
...
// and to retrieve:
byId[someId]

This has O(1) complexity. I can't imagine something more efficient without implementing it in native code (C++).

Our on-disk storage (which is engaged occasionally) is ACID compliant. Or in lame's terms, it is crash safe.

That said, there is always room for improvement (for example https://github.com/ArkEcosystem/core/blob/develop/packages/core-transaction-pool/src/mem.ts#L156), but any changes would better be well justified and tested in order to avoid performance regressions.

[fix]

We could store faster in-memory with reduced memory footprint by using aip11 and having getter/setter directly in the aip11 buffer.

For instance:

tx = new AIP11Transaction()
tx.setType(0)
tx.setRecipient(...)

The functions handling directly the data in buffer

This way we could store buffer in DB as well and

[faustbrian]

Ditching in favour of https://github.com/ArkEcosystem/core/issues/2114.