FILE/POS WITH SEMISYNC

Problem:
========
When DMLs are in progress on the master stopping a slave and
setting ahead binlog name/pos will cause an assert on the
master.

Analysis:
========
Trx1 is waiting for ack on certain position meanwhile change
master command is issued from the slave to master with a
future log file name and position. Upon receipt master's
dump thread assumes that it has received ack till the future
log file and position and clears the active transaction list
and sets the reply_file_name_ and reply_file_pos_ to future
file name and pos. since reply file name and position are
ahead of Trx1's current position Trx1 completes
successfully. Execute a new DML Trx2. During commit Trx2
waits for an ack on some position but the reply file name
and position are already pointing to future file name and
position so this current transaction thinks that it has
already got an ack and no need to wait and proceeds further.
But the active transaction has not actually received any ack
the transaction node is not cleared from active transaction
list. During the exit of commit operation there is an assert
which expects active transaction node to be cleared from the
transaction list which fails this leads to an assert.

Fix:
===
Ideally when a request for change master is received the
request file name and position should be validated for
correctness and only after the validation the transmit_start
hook should be fired.  Hence moved the transmit_start hook
position to appropriate place where it is placed post the
validation of file name and position. Post fix when a future
log file name and position is specified
ER_MASTER_FATAL_ERROR_READING_BINLOG error will be returned
to slave.

Note: During fixing this bug one more bug was identified.

Rpl_semi_sync_master_clients this parameter can become
negative at certain conditions causing UNINSTALL PLUGIN
rpl_semi_sync_master to fail.

'Rpl_semi_sync_master_clients' client is incremented when
dump thread initiates transmission and when a new slave
connection is established. This value is decremented when
dump thread exits and slave gets disconnected. On a certain
condition if IO thread stops due to an error and dump thread
exited before "transmit_start" hook dump thread will still
try to call "transmit_stop" and this will cause
Rpl_semi_sync_master_clients= -1 which is invalid state.
Hence as part of the fix a new flag is introduced such that
"transmit_stop" will be called only when "transmit_start" is
called.