docs(qwp): fix stale comments left behind by the Invariant-B budget-terminal removal (M4 residual)

Comment-only; no code changes. 6be20d6/e9a1e7d removed connectLoop's
wall-clock budget give-up and the never-connected/connection-lost
terminal, but four comments still described the old world:

- InitialConnectRetryTest.makeSfDir / InitialConnectAsyncTest.sfDirOpt
  claimed 'initial_connect_retry on/sync/async requires sf_dir per spec
  s3.5; without it the builder rejects construction' -- no such
  validation exists (Sender.build silently falls back to a memory-mode
  CursorSendEngine, and QuestDBBuilder.resolveLazyConnect injects
  initial_connect_retry=async with no sf_dir check). Now documents that
  the tests set sf_dir deliberately to exercise the durable SF path.
- InitialConnectAsyncTest class javadoc listed 'budget exhaustion' as a
  terminal delivered to the error inbox -- under Invariant B plain
  connect failures retry indefinitely; only auth / upgrade reject /
  durable-ack capability gap terminals reach the inbox.
- testWasEverConnectedTrueImmediatelyInSyncMode referenced the deleted
  never-connected-vs-connection-lost misclassification rationale.
- CursorWebSocketSendLoop.connectWithRetry javadoc said it is the 'same
  retry loop the I/O thread uses' -- the loops diverged: connectLoop
  retries forever (Invariant B), connectWithRetry (SYNC) remains bounded
  by maxDurationMillis and still throws on budget exhaustion.

PrReviewRedTestsE2e's budget-exhausted reference is left as-is: it is a
dated historical record of the PR-17 review findings.

Author: bluestreak01

core/src/main/java/io/questdb/client/cutlass/qwp/client/sf/cursor/CursorWebSocketSendLoop.java

@@ -374,11 +374,13 @@ public static SenderError.Category classify(byte status) {
     }
 
     /**
-     * Same retry-with-exponential-backoff-and-jitter loop the I/O thread
-     * uses on a wire failure, but reusable from {@code ensureConnected} to
-     * implement {@code initial_connect_retry=true}. Returns the connected
-     * client on success; throws on terminal upgrade error (won't retry) or
-     * budget exhaustion.
+     * Same exponential-backoff-with-jitter machinery as the I/O thread's
+     * {@code connectLoop}, but reusable from {@code ensureConnected} to
+     * implement {@code initial_connect_retry=true}. Unlike {@code connectLoop}
+     * (which retries indefinitely under Invariant B), this blocking variant
+     * IS bounded by {@code maxDurationMillis}: it returns the connected
+     * client on success and throws on terminal upgrade error (won't retry)
+     * or budget exhaustion.
      * <p>
      * Caller-supplied {@code factory} is invoked once per attempt and
      * should produce a fresh, connected, upgraded client (or throw). The

core/src/test/java/io/questdb/client/test/cutlass/qwp/client/InitialConnectAsyncTest.java

@@ -49,10 +49,11 @@
 /**
  * Behavior of {@code initial_connect_retry=async}: the producer-thread
  * {@code Sender.fromConfig} must return immediately even when no server
- * is reachable; the I/O thread retries connect in the background, and
- * terminal failures (auth/upgrade reject, budget exhaustion) are
- * delivered through the async error inbox rather than thrown at the
- * call site.
+ * is reachable; the I/O thread retries connect in the background. Plain
+ * connect failures are retried indefinitely (Invariant B: no wall-clock
+ * budget give-up); only genuine terminals (auth/upgrade reject,
+ * durable-ack capability gap) are delivered through the async error
+ * inbox rather than thrown at the call site.
  */
 public class InitialConnectAsyncTest {
 
@@ -283,8 +284,7 @@ public void testWasEverConnectedTrueImmediatelyInSyncMode() throws Exception {
         // before fromConfig returns. wasEverConnected() must therefore
         // already be true the instant the sender becomes visible to the
         // caller — there is no observable "never connected" window in
-        // those modes, so misclassifying a budget exhaustion as
-        // never-connected is impossible.
+        // those modes.
         try (TestWebSocketServer server = new TestWebSocketServer(new AckHandler())) {
             int port = server.getPort();
             server.start();
@@ -362,8 +362,11 @@ private static void assertCloseRethrowsTerminal(Sender sender, String expectedSu
 
     /**
      * Returns a unique temp sf_dir snippet for embedding in a config
-     * string. initial_connect_retry on/sync/async requires sf_dir per
-     * spec §3.5; without it the builder rejects construction.
+     * string. The builder does NOT require sf_dir for any
+     * initial_connect_retry mode — without it the sender builds in
+     * memory mode and buffers rows in the in-RAM cursor ring. These
+     * tests set an sf_dir so the rows accumulated before the first
+     * successful connect are disk-backed (the durable SF path).
      */
     private static String sfDirOpt() {
         String dir = java.nio.file.Paths.get(

core/src/test/java/io/questdb/client/test/cutlass/qwp/client/InitialConnectRetryTest.java

@@ -42,9 +42,12 @@
 public class InitialConnectRetryTest {
 
     /**
-     * Temp sf_dir for retry-mode tests. Per spec §3.5,
-     * initial_connect_retry on/sync/async requires sf_dir — memory-mode
-     * senders cannot durably retry across reconnects.
+     * Temp sf_dir for retry-mode tests. The builder does NOT require
+     * sf_dir for any initial_connect_retry mode — memory-mode senders
+     * share the same retry machinery, buffering rows in the in-RAM
+     * cursor ring instead of on disk. These tests use an sf_dir so the
+     * retried rows are disk-backed and the tests exercise the durable
+     * SF path.
      */
     private static String makeSfDir() {
         return java.nio.file.Paths.get(