Add subscription drop/resubscribe tests

[alexeyzimarev]

For all the available stores (relational and ESDB) add tests to ensure that the subscription drop causes resubscribe.

Also ensure that the health check returns degraded when the subscription is dropped.

I've seen that resubscribe logic works fine for ESDB as when I run the sample app and stop the ESDB container, it enters the loop of doom trying to resubscribe. As soon as the container starts again, the subscription restores. For SQL Server, #307 indicates that it doesn't resubscribe, so it's best to start with adding a test there.

[linear]

EVE-49 Add subscription drop/resubscribe tests

[alexeyzimarev]

Added automated drop/resubscribe coverage following the existing HandlerFailureBase convention: one infra-agnostic base test plus a thin per-store subclass.

SubscriptionDropBase (in Eventuous.Tests.Subscriptions.Base) exercises the full cycle and covers both asks from this issue in a single flow:

1. Produce and consume events, assert the subscription health check reports Healthy.
2. Simulate a transport drop by pausing the infrastructure container (docker pause, which freezes the DB process but keeps the published port, so the same connection string keeps working after recovery).
3. Wait for the subscription to detect the drop, assert health flips to Unhealthy.
4. Unpause the container.
5. Wait for the subscription to resubscribe, assert health is back to Healthy.
6. Produce more events and assert they get processed — so recovery is verified functionally, not just by the reported state.

Wired into the container-backed stores via the existing subscription fixtures, all passing locally:

• SQL Server — SqlServer_ShouldResubscribeAfterConnectionDrop
• PostgreSQL — Postgres_ShouldResubscribeAfterConnectionDrop
• KurrentDB (ESDB) — Esdb_ShouldResubscribeAfterConnectionDrop, with [Retry(3)] matching its handler-failure sibling

SQLite is intentionally excluded — its fixture isn't container-backed, so there's no transport to drop.

Notes:

• The SQL Server case (the Transient error drops the subscription. (Connection reset by peer) #307 concern) passes — relational subscriptions do call Dropped(DropReason.ServerError, …) on a non-transient connection failure and resubscribe correctly in the current code.
• On the health check wording: SubscriptionHealthCheck only exposes Healthy/Unhealthy (there is no Degraded status), so the test asserts Unhealthy while dropped and Healthy after recovery. The fixture wires the subscribed/dropped callbacks to the health check exactly the way SubscriptionHostedService does in production.
• These tests cover the $all (all-stream) subscriptions; stream-scoped subscriptions can get the same coverage trivially if we want it.

With this in place the original gaps — "subscription drop causes resubscribe" and "health check reports unhealthy when dropped" — are covered for all three container stores.