Well, Zabbix's database connector does not use DNS SRV records to locate the database. According to the manual, you need to provide the exact connection details.
Using haproxy to make database reliable should be possible.
As I see you are trying to build highly available solution here, make sure you've read the Zabbix's own documentation regarding that. Also "MySQL cluster" might mean different things; if you meant Galera cluster then I'm afraid that wouldn't be the best possible solution. The Zabbix database usage pattern is to write many small records into few tables (one table per handled data type), which might be not the best usage pattern for the Galera. The standard primary-standby (or "master-slave") database replication is the best match to the Zabbix's built-in HA capability.