[FEA] BSDG: Add Stratify option

BSDG (Billion Scale Data Generator)
Open-sourced version in cuvs bench currently generates the synthetic dataset cluster by cluster, and returns/saves dataset in that order.
Add a stratify option so users can get a shuffled version of the dataset.

We may need to redundantly generate each cluster multiple times to do this.

Existing Method:

• Generate cluster 0 (N_0 vectors) -> write to disk
• Generate cluster 1 (N_1 vectors) -> write to disk
  ...

Stratify Option (example):

data_len = 0
while data_len < N:
    for c in clusters:
        c_data = gen_cluster(c)
        f.flush(pick_new_rows(c_data, c)) # we just grab a few rows that we haven't added to the data so far

Start by locating BSDG in the cuvs bench code and reading the existing cluster-by-cluster generation and dataset-writing flow. The work is done when a stratify option produces the requested shuffled dataset without changing the generated row count; the issue does not name a test or specific file to run.