Filter out the downloaded sample columns and only store the custody requirement columns

[jimmygchen]

Description

The initial version of subnet sampling for PeerDAS was implemented in #6410, it allows for interop testing but does not filter out the sampled columns, ie. all downloaded columns, including columns for both custody and sampling are stored - however we only need to store the custody requirement columns, as peers should never query the node for columns that the node is not assigned to custody.

The change wasn't made in #6410 because it requires a bit more work, and also creates more work to switch back to Peer Sampling. However, If we do go live with Subnet Sampling, then it makes sense to exclude storing the sampling columns to save storage for non validating full nodes (a validating node will custody at least 8 columns, so it makes no difference). However given that PeerDAS has been postponed to the fork after PectraA, it probably make sense to hold off until then - as we could end up implementing Peer Sampling.