fix(api): report admission saturation in /api/health without marking the proxy dead

[ANAMASGARD]

Summary

Fixes #244

When the proxy hits its max active downstream limit, it’s still healthy — it’s just not accepting new miners. Before this change, /api/health could return 503 based on signals that don’t reflect real admission capacity (handoff queue depth or stats count). Load balancers and operators could treat a live, working proxy as dead and keep sending miners to it anyway.

This PR separates liveness (“is the proxy up?”) from readiness (“can it take more miners?”):

• 503 only when something is actually wrong: translator handoff channel is closed, or the proxy is globally down.
• 200 when the proxy is alive, with a JSON body that reports saturation clearly.

Saturation is read from the ingress admission semaphore — the same gate that stops new TCP accepts when the proxy is full — not from the handoff queue (which drains quickly into background setup) or from stats.len() (which can lag).

What changed

Before	After
Full handoff queue → 503	200 + handoff_backlog_full: true
stats.len() >= max → 503	200 + saturated: true (from semaphore)
Alive and healthy → 200 "Proxy OK"	200 with structured body

Example response when alive but at capacity:

{
  "success": true,
  "data": {
    "status": "ok",
    "saturated": true,
    "handoff_backlog_full": false,
    "message": "Proxy is alive but admission saturated: not accepting new connections"
  }
}