Memory leak: New subtensor connection created on each weight setting instead of reusing existing one

[rusvo]

Description

In bettensor/validator/utils/scoring/weights_functions.py, a new subtensor connection is created every time weights are set:

weights_subtensor = bt.subtensor(network=self.neuron_config.network)This causes:

• WebSocket connection leaks (connections not properly closed)
• Memory leak (each connection holds buffers)
• "Task was destroyed but it is pending!" warnings in async-substrate-interface
• High memory usage over time (observed 2.4GB+ on validator)

Suggested Fix

Reuse the existing subtensor connection:

weights_subtensor = self.subtensor## Environment

• bettensor version: main branch (commit fbcb055)
• async-substrate-interface: 1.5.9
• Python: 3.x
• OS: Ubuntu

Additional Context

There may be other places in the codebase with similar issues where bt.subtensor() is called instead of reusing an existing connection.