Problem Description:
In Zabbix 7.0, the current Proxy Group load balancing algorithm distributes hosts evenly across proxies based solely on host count, but completely ignores the actual monitoring workload measured by Number of Values Per Second. This causes severe imbalance in large-scale environments.  

Specific Scenario:  

  Server: Zabbix Server 7.0.11

  Proxy: Zabbix Proxy 7.0.11

  Total Hosts: 20,000+   

  Proxy Group: 4 proxies

   Host Distribution:  
     10,000+ hosts: ~70 items per host (e.g., SNMP metrics, HTTP checks, log monitoring).           10,000+ hosts: Only 3 basic items (e.g., ICMP ping, TCP port checks).  

Result:  
  Proxies handling high-NVPS hosts reach 900-1000 NVPS(overloaded).  
  Proxies handling low-NVPS hosts run at 30-40 NVPS(underutilized).  

Critical Issue: Despite "balanced" host counts per proxy (~5,000 hosts each), the actual monitoring load differs by 20-30x.  

Current Behavior
  Proxy assignment logic: `Total Hosts ÷ Number of Proxies` = ~5,000 hosts per proxy.  
  No consideration for:  
  - Total items per host  
  - Estimated NVPS per proxy  
  - Proxy hardware capacity differences