Problem:

When dealing with large GTID sets (bigger than 64K bytes when converted
to string), the "performance_schema.replication_connection_status" is
truncating the "RECEIVED_TRANSACTION_SET".

Analysis:

The "RECEIVED_TRANSACTION_SET" is created of type TEXT, and supports
strings up to 64K bytes of size.

When dealing with large/complex clusters of servers using replication
it is possible that sometimes the GTID sets to handle have lots of
holes, leading to very large GTID set strings.

When trying to certify if a receiver thread has received all the GTIDs
a master has executed, we can compare master's GTID_EXECUTED with the
slave's channel "performance_schema.replication_connection_status"
"RETRIEVED_GTID_SET" field. But because of the limitation of the max
string length of a TEXT field type, this information will be truncated
making a comparison to assume that the slave didn't received all the
master's GTIDs yet.

Fix:

Changed the type of "RECEIVED_TRANSACTION_SET" field at
"performance_schema.replication_connection_status" table from "TEXT"
to "LONGTEXT".

While fixing the issue, we also changed some Gtid_set functions return
types and variable types from int to size_t or long in order to avoid
future issues with the GTID set conversion to strings.