Allow flush during translog replay

When replaying from translog flushes are currently not allowed for
historic reasons in stateful. However, the way translog is handled
in stateless reintroduced a similar reason - when we flush we assume
that the newest translog is all that need to be replayed in case of
a crash. Fix this to now copy over the translog info if we flush
during translog replay.

Flushes can primarily happen during translog replay due to
inaccuracies in the live version map causing it to go unsafe,
subsequently causing a refresh, which converts toa flush in
stateless. However, we could run in a disk constrained env during
replay and as such allowing flushes seems safer.

server/src/main/java/org/elasticsearch/index/engine/InternalEngine.java

@@ -632,14 +632,22 @@ private void recoverFromTranslogInternal(
             // thread to continue with recovery, but if it doesn't do anything async then there's no need to fork, hence why we use a
             // SubscribableListener here
             final var flushListener = new SubscribableListener<FlushResult>();
-            flush(false, true, flushListener);
+            flushCompletedTranslogRecovery(flushListener);
             flushListener.addListener(l.delegateFailureAndWrap((ll, r) -> {
                 translog.trimUnreferencedReaders();
                 ll.onResponse(null);
             }), engineConfig.getThreadPool().generic(), null);
         });
     }
 
+    protected void flushCompletedTranslogRecovery(SubscribableListener<FlushResult> flushListener) {
+        flush(false, true, flushListener);
+    }
+
+    protected boolean pendingTranslogRecovery() {
+        return pendingTranslogRecovery.get();
+    }
+
     protected Translog.Snapshot newTranslogSnapshot(long fromSeqNo, long toSeqNo) throws IOException {
         return translog.newSnapshot(fromSeqNo, toSeqNo);
     }
@@ -2979,7 +2987,9 @@ protected Map<String, String> getCommitExtraUserData() {
         return Collections.emptyMap();
     }
 
-    final void ensureCanFlush() {
+    // todo: this protection is no longer needed, after ES now relies on sequence numbers for figuring out which translog to replay.
+    // in the past, ES would attach the latest translog generation to a commit, making it unsafe to flush during translog replay.
+    protected void ensureCanFlush() {
         // translog recovery happens after the engine is fully constructed.
         // If we are in this stage we have to prevent flushes from this
         // engine otherwise we might loose documents if the flush succeeds