feat(spooler): Add EnvelopeStack based on SQLite

[iambriccardo]

This PR adds the SQLite implementation of the EnvelopeStack trait. The implementation works in the following way:

• It keeps a buffer of envelopes in memory, which comprises a stack of arrays, each with a fixed size.
• The buffer is always filled from the top and flushed to disk only when the size surpasses a certain threshold.
• When reading, if the buffer is empty, a load from the disk will be attempted.

The idea of the implementation is to allow an in-memory data structure to always serve the most recent envelope and only go to disk when really necessary. So, during low load, it could easily be that the memory buffer is only used and never spilled to disk (assuming thresholds are defined correctly).

It's important to note that we are trading off durability for speed with this design.

Closes: #3855

[jjbayer]

Looks good, I think we can merge this into master soon, and then decide on final interfaces (sync vs async etc.) once I merge it with my work.

[jjbayer]

I'm a bit confused about pushing batches to the front but individual elements to the back. Can we sort both the batches_buffer and the individual batches left-to-right, i.e. push to the back in both cases?

[jjbayer]

As discussed, push front vs push back.