finalize_all_subnet_root_dividends has unbounded iteration over all hotkeys

[ppolewicz]

In pallets/subtensor/src/staking/claim_root.rs, the finalize_all_subnet_root_dividends function collects ALL hotkeys from RootClaimable storage into memory, then mutates each one:

pub fn finalize_all_subnet_root_dividends(netuid: NetUid) {
    let hotkeys = RootClaimable::<T>::iter_keys().collect::<Vec<_>>();
    for hotkey in hotkeys.iter() {
        RootClaimable::<T>::mutate(hotkey, |claimable| {
            claimable.remove(&netuid);
        });
    }
    let _ = RootClaimed::<T>::clear_prefix((netuid,), u32::MAX, None);
}

This is O(N) reads + O(N) writes where N is the total number of hotkeys with any root claimable balance. This function is called from do_dissolve_network, which is an extrinsic. With thousands of hotkeys, this could:

• Exceed block weight limits
• Cause high memory usage from the collect::<Vec<_>>()

The weight annotation on do_dissolve_network may not account for this unbounded iteration.

Suggested fix: Consider using RootClaimable::iter_keys() without collecting, or implementing a bounded/paginated cleanup, or using a different storage layout that allows per-netuid cleanup without iterating all hotkeys.

[open-junius]

do_dissolve_network needs much more weights than it is defined in the extrinsic. Will investigate how to do it with fixed weight consumption.