Use GTID instead of filename/pos for semisync ack also for legacy binlog

Following MDEV-38190, the semi-sync will use GTID for identifying transactions in acknowledement with binlog-in-engine. But the legacy binlog still uses filename/offset as identifiers for acks.

This issue is for the idea to eventually switch to use GTID in both places.

There isn't really any benefit to using GTID over filename/offset (in the legacy binlog case), it's just an opaque identifier now that's sent back and forth. Ideally we would only use GTID for consistency, but the filename/offset needs to be supported for backwards compatibility anyway.

Mailing list thread: https://lists.mariadb.org/hyperkitty/list/developers@lists.mariadb.org/thread/IDMHMOYOFJWHB5DP6W3S3Y3K7GYMVVAC/