Review request: retry failed recovery notifications (#1833)

[WyattJia]

Hi @TwiN, could you take another look at PR #1833 when you have a chance? It fixes lost recovery notifications in the alerting path: a temporary provider error currently clears the pending recovery, so later healthy checks never retry it. The patch retains pending state until delivery succeeds and retries through the existing evaluation cycle.

I have addressed the review follow-ups with clearer state-transition comments, SQLite restart coverage, and per-endpoint serialization for concurrent external pushes and heartbeat checks.

Validation went beyond unit tests: I ran the actual race-enabled Gatus executable against a local HTTP receiver and SQLite. In controlled runs, nine concurrent healthy pushes produced one recovery notification instead of nine; new failures and heartbeat expiry were preserved; and pending recovery survived a process restart and was retried. All 24 related packages passed with -race, and four concurrent API scenarios passed ten repetitions. Detailed follow-up and results.

I would appreciate a maintainer review and guidance on anything still needed before merging. Happy to make further changes. Thanks!