Investigate RTT effect on the num txs per block variance

When we migrate to smaller slots of 200ms and later to 2slots leader window, client side will be more sensitive to the latency of receiving slot updates (~50ms) and latency of sending transactions (mean is ~50ms for client Amsterdam, but might be 200ms for Tokyo).
The current heuristic in tpu-client-next is to send tx to both current and next leader when we are at the last slot in window. But this will be not enough.

Start with the current tpu-client-next last-slot heuristic and compare it with the stated 200ms slots, 2-slot leader window, and 50–200ms latencies. Measure or model how RTT affects transactions-per-block variance. Done means a quantified result and a clear conclusion about whether sending to the current and next leader is sufficient.

Written by the indexing model from the issue text.