samples uids only on candidate uids

[mccrindlebrian]

The current implementation of uid sampling is very weird, and I don't understand it. Currently, if the number of uids is < the number of samples we want in a batch (k) (which is rare), we will take all the samples in the network, and then sample uids that are NOT in the candidate uids AND in the avail_uids variable. Therefore, this samples uids that are explicitly excluded from from the original set of candidate uids. Unless I am mis-reading this, it seems like a very strange implementation to me. The reasoning for this, for what I can tell, is to sample uids until we reach k, even if the uids are in the exclusion category.

I don't like this implementation, and it requires the validator to set a value of k that they won't have prior knowledge to beforehand. In production, this isn't an issue, since we will likely always have candidate_uids >> k. However, in testnet scenarios, you have no idea. I ran into many issues regarding sampling when trying to deploy a validator on testnet because of the logic of sampling.

Therefore, I propose a different uid sampling procedure, where we do not artificially want the batch to be 50, but rather the batch is always capped to be the max number of candidate uids if candidate_uids < k. Otherwise, we randomly sample ONLY from the list of candidate_uids. This removes the need for the available_uids list.