Fix leafnode reconnect stall after compressed write timeout

After a partial write, the pending-byte count includes both leftover
compressed data and new data waiting to be compressed. The compression
code subtracted that entire count when replacing the new data with its
compressed form, losing track of the leftovers. This could cause the
write counts to go negative, which would stall the connection.

Signed-off-by: Neil Twigg <neil@nats.io>