NTCP write should be in a loop

[zab]

This is symmetric to #2243 except for writing. There is no guarantee that a single write call will write the entire buffer, so the write should be in a loop until it returns 0 or there is nothing more to write.

One possible way of doing this is:

#
# old_revision [5175d0f5f52c62ecc0e2e76c53e445d47181ec9e]
#
# patch "router/java/src/net/i2p/router/transport/ntcp/EventPumper.java"
#  from [f284aa0bf8e6a8fe2b465e349d98e677a5939021]
#    to [94a79661753d0228e6ec34d8cf909ca6f71c876b]
#
============================================================
--- router/java/src/net/i2p/router/transport/ntcp/EventPumper.java      f284aa0bf8e6a8fe2b465e349d98e677a5939021
+++ router/java/src/net/i2p/router/transport/ntcp/EventPumper.java      94a79661753d0228e6ec34d8cf909ca6f71c876b
@@ -753,7 +749,7 @@ class EventPumper implements Runnable {
                     } else if (buf.remaining() > 0) {
                         // stay interested
                         //key.interestOps(key.interestOps() | SelectionKey.OP_WRITE);
-                        break;
+                        continue;
                     } else {
                         con.removeWriteBuf(buf);
                         //if (buffer time is too much, add OP_WRITe to the interest ops and break?)

[zzz]

agreed

[zzz]

In 9c206a80263cb001f5edf53dbb383b3d48aa22b3 to be 0.9.39-10