feat(backend): stream cluster stats over SSE

[a9ex]

Description

Briefly describe what this PR does :

Reworks the cluster stats SSE routes to consume Docker's native stats stream instead of polling.

• GET /clusters/:id/stats/stream now holds a single persistent Docker stats stream and forwards each sample as it arrives, instead of re-requesting stats on a timer. Interval query param is dropped for this route (Docker drives the cadence).

• GET /clusters/stats/stream keeps one persistent stats stream open per running container, caches the latest sample per cluster, and emits the merged snapshot every N seconds. Clusters changes (started/stopped) is reconciled on the same cadence, and streams that end on their own are dropped.

• Bumps @hallmaster/docker.js to 0.0.28 (the version exposing the streaming overload).

---

Context / Motivation

Explain why this change is needed:

The stats SSE routes polled Docker with stream=false on a timer. In that mode Docker "waits for 2 cycles" to compute the CPU delta before returning a single sample and disconnecting.
With exhaustMap, whenever a cycle overran the interval the following ticks were dropped, so at short intervals (e.g. 1s over several containers) the daemon couldn't keep up and the SSE cadence became erratic.

---

Related Links

HallMasterOrg/docker.js#20

---

Screenshots (if applicable)

N/A, video too big

---

Checklist

• My code follows the project's coding style
• Tests pass locally
• I added/updated relevant tests
• I updated documentation (if needed)
• This PR is ready for review

---

Additional Notes