fix(deploy): notify on container crash loop after deployment

[AminDhouib]

Ports Dokploy#4357 by @sajdakabir (1 commit, cherry-picked with -x, original authorship preserved) + 1 fork-side test-adaptation commit.

Addresses Dokploy#4339 (verified: linked via the upstream PR's closing references; the diff makes a post-deploy crash loop surface as a failed deployment + build-error notification instead of a silent "done").

What / why

Previously a deployment was marked done (and success notifications sent) as soon as the container was (re)created — if the app then crash-looped, nothing was reported.

• New waitForSwarmServiceStable(appName, { serverId, windowMs = 60s, pollMs = 5s }) in packages/server/src/utils/docker/utils.ts: polls the swarm service's tasks; latest task failed/rejected, or a restart after having been running, → unstable (with reason).
• deployApplication / rebuildApplication run this check after mechanizeDockerContainer; instability throws, which routes into the existing error path: deployment/application marked error and sendBuildErrorNotifications fires (existing payloads only — no DB schema change, no migration).

Reviewer heads-up: by upstream design, every successful application deploy now includes a fixed 60s stability window before being marked done.

Adaptations (fork-side, test-only)

Upstream PR Dokploy#4357's own pr-check (test) is red; reproduced locally — the stability poll hits a real Docker socket from unit tests:

1. __test__/deploy/application.command.test.ts: partial mock of docker/utils stubbing waitForSwarmServiceStable → { stable: true } (rest of the module stays real). Without it, 4 command-generation tests time out (5s limit vs 60s poll) — exactly upstream's CI failure.
2. __test__/deploy/application.real.test.ts: REAL_TEST_TIMEOUT 180s → 300s, since each successful real deploy now includes the fixed 60s window on top of real nixpacks/railpack build time.

No production-code deviation from upstream.

Test evidence (Windows local; Linux CI is source of truth)

• __test__/deploy/ suite: 52 passed / 1 skipped; application.command.test.ts 6/6 (the 4 pre-adaptation timeouts are gone). Only application.real.test.ts fails locally — in beforeEach shell setup, pre-existing Windows-local behavior for this Linux-only file, unrelated to this change.
• pnpm --filter=@dokploy/server run typecheck and pnpm --filter=dokploy run typecheck: clean.

Sibling / merge-order notes

• Port of upstream fix: prevent SSH private key from leaking into build-failure notifications Dokploy/dokploy#4630 (fork PR fix: prevent SSH private key from leaking into build-failure notifications #63) also touches packages/server/src/services/application.ts. Hunk analysis: fix: prevent SSH private key from leaking into build-failure notifications #63 edits the catch block (~L250) and the ./deployment import; this PR edits the try block (~L222/L323) and the docker/utils import — change-disjoint, either merge order is clean; re-run CI on whichever merges second. Once both land, a crash-loop failure's notification message flows through fix: prevent SSH private key from leaking into build-failure notifications #63's log-tail sanitizer, which is the desired combined behavior.
• No file overlap with open fork PRs feat(security): display a masked webhook url by default with optional toggle #36/fix(notifications): use real server name in remote threshold alerts #58.