docs(audience): drop stale _consent refs and tighten _state lock comment

Two comments still named the pre-refactor _consent field; one class-level
comment enumerated only Identify / Reset as taking _initLock, which is
now true of every _state write.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: ImmutableJeffrey

src/Packages/Audience/Runtime/ImmutableAudience.cs

@@ -15,9 +15,9 @@ public static class ImmutableAudience
         // Reference fields are written inside _initLock; readers check the
         // `volatile _initialized` flag first so they never see a half-initialised state.
         // _state (consent level + userId) and _session are volatile so a write
-        // on one thread is visible on any other. Writes happen under _initLock
-        // (Identify / Reset / SetConsent also take it) so level and userId
-        // always move together — callers never observe (Anonymous, oldUserId).
+        // on one thread is visible on any other. Every _state write happens
+        // under _initLock so level and userId always move together — callers
+        // never observe (Anonymous, oldUserId).
         //
         // Init / Shutdown / Reset / SetConsent hold _initLock only to flip state
         // and capture references; they release the lock before running blocking
@@ -789,7 +789,7 @@ private static void RescheduleSendTimer(HttpTransport transport)
             timer.Change(nextMs, sendIntervalMs);
         }
 
-        // consentAtInit only gates the launch; Track still checks live _consent via CanTrack.
+        // consentAtInit only gates the launch; Track still checks live _state via CanTrack.
         private static void FireGameLaunch(AudienceConfig config, ConsentLevel consentAtInit)
         {
             if (!consentAtInit.CanTrack()) return;

src/Packages/Audience/Tests/Runtime/ImmutableAudienceTests.cs

@@ -568,7 +568,7 @@ public void SetConsent_DowngradeToNone_StressTest_NoLeak()
             // before SetConsent starts on a fast machine. This stress variant runs
             // the race many times with many concurrent Track threads so at least
             // some iterations are guaranteed to land the enqueue inside the
-            // _consent=None/PurgeAll window.
+            // _state=None/PurgeAll window.
             //
             // Without the EnqueueChecked re-check, this test leaks events
             // reproducibly. With the fix, zero leaks across all iterations.