Refactor to polling-first architecture with OBSERVE as accelerator

State freshness now comes from a tiered PollScheduler over the persistent
DTLS session; OBSERVE registrations are kept as an opportunistic
acceleration layer. Behaviour is identical online vs air-gapped except
for worst-case freshness latency.

Adds three modules:
- StateCache: single source of truth, source-tagged change events
- PollScheduler: hot/warm/cold + sweep tiers, write-defer past the
  fetchback-revert window, per-window RTT/slow-poll tracking
- KeepaliveTask: CoAP empty-CON ping with consecutive-fail detection
  driving MQTT availability

Bridge publishes per-appliance diagnostic entities (Push Active, Last
Update Source, Poll Max RTT, Slow Polls, Poll Errors, Stalest Resource
Age, Last OBSERVE Age) under HA's Diagnostic section. Tier cadences
are descriptor-declared, calibrated against measured per-firmware
ceilings (dryer ~14 req/s, oven ~8 req/s via probe_poll_rate_combined.py).

Drops HEARTBEAT_INTERVAL_S in favour of the descriptor-declared sweep
tier; PING_INTERVAL_S now consumed by KeepaliveTask inside the bridge
rather than driven from main.py.

README explains the push/poll split and what happens when the appliance
is blocked from internet.

samsung_appliance/state_cache.py

@@ -0,0 +1,78 @@
+"""Single source of truth for one appliance's state.
+
+All writers (OBSERVE notify, poll, seed, optimistic) call apply_rep().
+A registered on_change callback fires after any apply that mutated the
+cache, which the bridge wires to its MQTT publish gate.
+"""
+from __future__ import annotations
+
+import threading
+import time
+from typing import Callable, Optional, TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from .appliances.base import ApplianceDescriptor
+
+
+class StateCache:
+
+    def __init__(self, descriptor: 'ApplianceDescriptor'):
+        self.descriptor = descriptor
+        self.links: dict[str, dict] = {}
+        self.last_updated: dict[str, float] = {}
+        self.source: dict[str, str] = {}
+        self.descriptor_state: dict = {}
+        self._on_change: Optional[Callable[[bool, str], None]] = None
+        self._lock = threading.RLock()
+
+    def set_on_change(self, cb: Callable[[bool, str], None]) -> None:
+        self._on_change = cb
+
+    def apply_rep(self, href: str, rep: dict, source: str) -> bool:
+        if not isinstance(rep, dict):
+            return False
+        with self._lock:
+            prior = self.links.get(href)
+            changed = prior != rep
+            self.links[href] = rep
+            self.last_updated[href] = time.time()
+            self.source[href] = source
+        hook = self.descriptor.on_observation
+        if hook is not None:
+            try:
+                hook(self.descriptor_state, href, rep)
+            except Exception:
+                pass
+        if self._on_change is not None:
+            try:
+                self._on_change(changed, source)
+            except Exception:
+                pass
+        return changed
+
+    def apply_optimistic(self, href: str, body: dict) -> bool:
+        if not isinstance(body, dict):
+            return False
+        with self._lock:
+            merged = dict(self.links.get(href) or {})
+            merged.update(body)
+        return self.apply_rep(href, merged, source='optimistic')
+
+    def get(self, href: str) -> Optional[dict]:
+        with self._lock:
+            return self.links.get(href)
+
+    def snapshot(self) -> dict[str, dict]:
+        with self._lock:
+            return dict(self.links)
+
+    def freshness_s(self, href: str) -> Optional[float]:
+        ts = self.last_updated.get(href)
+        return None if ts is None else (time.time() - ts)
+
+    def stalest(self) -> Optional[tuple[str, float]]:
+        with self._lock:
+            if not self.last_updated:
+                return None
+            href = min(self.last_updated, key=self.last_updated.get)
+            return href, time.time() - self.last_updated[href]