[Darwin] Coalesce mDNS resolve cancel/restart so in-flight results aren't dropped

User-visible failure: on Darwin, every reconnect to a Matter node after
ChipDnssdResolveNoLongerNeeded shows ~1s of extra NodeID-resolve latency.
Inbound mDNS resolve answers already queued on the dnssd socket get discarded,
so the next resolve has to start from scratch.

Root cause: when the consumer counter drops to zero we immediately call
Finalize -> DNSServiceRefDeallocate. Per the dnssd contract,
DNSServiceRefDeallocate discards any events queued on that connection but not
yet read. A second observation from the mDNS owner is that "starting and
stopping queries doesn't query harder" -- a tight cancel-then-restart for the
same instance name is strictly worse than letting the existing query run.

Fix: introduce a per-ResolveContext deferred-teardown window (default 500ms)
before the actual DNSServiceRefDeallocate. Inside the window: a queued read
indicator dispatches the result through DispatchSuccess (which cancels the
timer); a new ChipDnssdResolve for the same instance name reuses the existing
context and bumps the counter back to 1, skipping DNSServiceCreateConnection /
DNSServiceResolve entirely; otherwise the timer fires
OnResolveDeferredTeardown -> Finalize(CHIP_ERROR_CANCELLED), preserving the
existing failure-path contract upper layers rely on.

Carve-out: delegate-based ResolveContexts (callback == nullptr, used by
MTRCommissionableBrowser) are NOT subject to deferred teardown. The browser
churns OnBrowseAdd/OnBrowseRemove for the same instance name on the order of
microseconds while a device is being discovered; holding the underlying
DNSServiceRef alive across that churn starves DNSServiceGetAddrInfo of a
chance to deliver before the next remove arrives, which manifested as
MTRCommissionableBrowserTests/test005 timing out under TSAN. The NodeID
reconnect bug this PR fixes is on the callback-based path.

Blast radius is confined to the Darwin dnssd platform layer.

Tests in src/platform/tests/TestDnssd.cpp pin:
- ReusesContextWithinDeferredWindow (callback-based coalescing)
- DelegateBasedResolveIsNotDeferred (delegate-based synchronous teardown)
- CancelStillPropagatesIfNoInFlightResult (timer fires once if no follow-up)
- Multi-sibling rescue, scope mismatch refusal, mismatched-callback rebind,
  shared-counter invariants, repeated toggle within window, etc.

rdar://176263876

Author: woody-apple

src/platform/Darwin/dnssd/DnssdContexts.cpp

@@ -525,13 +525,27 @@ ResolveContext::ResolveContext(DiscoverNodeDelegate * delegate, chip::Inet::IPAd
 void ResolveContext::DispatchFailure(const char * errorStr, CHIP_ERROR err)
 {
     ChipLogError(Discovery, "Mdns: Resolve failure (%s)", errorStr);
+    // If a deferred teardown timer was scheduled against us, cancel it so it
+    // doesn't fire after we are deleted.
+    CancelDeferredTeardownIfScheduled(this);
     // Remove before dispatching, so calls back into
     // ChipDnssdResolveNoLongerNeeded don't find us and try to also remove us.
     bool needDelete = MdnsContexts::GetInstance().RemoveWithoutDeleting(this);
 
     if (nullptr == callback)
     {
-        // Nothing to do.
+        // Delegate-based ResolveContext (no DnssdResolveCallback). The
+        // deferred-teardown rescue/coalesce path can dispatch
+        // CHIP_ERROR_CANCELLED here long after the consumer thinks the
+        // resolve has gone away, but the delegate API has no failure-signal
+        // hook. Log so the silent drop is at least observable in the field.
+        if (err != CHIP_NO_ERROR)
+        {
+            ChipLogError(Discovery,
+                         "Mdns: Delegate-based resolve %s failed with %" CHIP_ERROR_FORMAT
+                         " -- no delegate failure callback exists; silently dropping",
+                         instanceName.c_str(), err.Format());
+        }
     }
     else
     {
@@ -546,6 +560,9 @@ void ResolveContext::DispatchFailure(const char * errorStr, CHIP_ERROR err)
 
 void ResolveContext::DispatchSuccess()
 {
+    // If a deferred teardown timer was scheduled against us, cancel it so it
+    // doesn't fire after we are deleted.
+    CancelDeferredTeardownIfScheduled(this);
     // Remove before dispatching, so calls back into
     // ChipDnssdResolveNoLongerNeeded don't find us and try to also remove us.
     bool needDelete = MdnsContexts::GetInstance().RemoveWithoutDeleting(this);