GTID AND LOW SLAVE_NET_TIMEOUTS

Problem: When GTID is enabled, dump thread is not checking
the necessity of heartbeat event while it is scanning
through the binary log files and skipping some GTID groups
which were already present at Slave.

Analysis: Dump thread sends a heartbeat event to Slave if
there are no events to send for "heartbeat_period" seconds
to make the connection between Master and Slave active.
But when dump thread is scanning a binary log file and
if it finds many GTID groups(/events) that needs to be skipped,
it is not looking for this time period and not
looking to send heartbeat event to Slave.

There are two problems with the existing code in this
scenario:

Problem 1: If dump thread is spending more time in skipping the groups
(many groups that needs to be skipped) and is not sending any
heartbeat event, Slave thinks that Master is dead and it will
try to reestablish the connection.
Problem 2): Dump thread has two while loops to process the events at Master
side, a ) outer loop: to process all binary log files one by one
      b ) inner loop: to process all the events one by one in a file

Outer loop is having a flag 'thd->killed' to check if dump thread is killed
in between processing the different files, if so, it exists the while loop.

But Inner loop is not having any checks like this which end up in processing
the full binary file( if it is huge, taking more time) which is unnecessary
if this dump thread is killed due to some reason (One reason could be that
this dump thread could have been detected as Zombie thread by another new
dump request from Slave).

Fix:
1) Dump thread will now check whether it is time to send an heartbeat event
before skipping an event. If so, it will send one heartbeat event to Slave.
2) Inner loop also checks for thd->killed flag to avoid unnecessary work.