feat(spooler): Implement shutdown behavior in the spooler

[iambriccardo]

This PR implements the shutdown behavior in the spooler which works in the following way:

1. The shutdown service notifies the buffer service of shutting down.
2. If the shutdown signal has a timeout, the buffer will try to spool to disk all the envelopes that it has in memory. Other implementations of the envelope stack will just drop envelopes.
3. The spooling will take place in batch sizes of arbitrary size (100 for now), meaning that we will insert at most 100 elements per insertion. This is done so that we have more granular insertion in case of failures.
4. In case the shutdown takes more than expected, the future will be cancelled and an error will be shown to the user.

Note that cancellation of the shutdown procedure might lead to data loss based on the progress w.r.t. the await points.

Closes: https://github.com/getsentry/team-ingest/issues/529

[iambriccardo]

The big question I have for this PR, is which strategy should we employ to deal with the envelopes in the stacks? To me, the strategy depends on the implementation:

• Memory -> we drain and send all envelopes to the project cache (within the time bounds, if any).
• Disk -> we drain and persist all envelopes on disk (within the time bounds, if any).

In general, given the shutdown time bound, we still don't have to guarantee full persistence of envelopes but we can do a best effort.

[iambriccardo]

Decided to branch out here and not call any flush method on the memory impl, since I didn't want to make the flushing empty in the provider or stack, since it made no sense.

[jjbayer]

Excellent choice!

nit: I would prefer an explicit match on all variants here. If we ever add, let's say, a CloudStorageEnvelopeBuffer, the compiler would then force us to implement the new match arm.

[iambriccardo]

Yes, that's a good point. I initially implemented it that way, but felt it was too verbose. But your argument makes it a more valid approach! Will fix

[jjbayer]

Excellent choice!

nit: I would prefer an explicit match on all variants here. If we ever add, let's say, a CloudStorageEnvelopeBuffer, the compiler would then force us to implement the new match arm.