`DesignSpace.estimate_cost()` only considers the cost of first sim in the batch

[tomflexcompute]

DesignSpace.estimate_cost() gives an incorrect cost estimate since it seems like each individual task's cost is given only by the first sim in the sweep. Then the total cost is just the cost of the first sim * number of sims, which could largely over- or under-estimate the total cost for the batch since the individual task costs can vary a lot within the batch.

[tomflexcompute]

Related issue: #2506

[tylerflex]

I think actually it looks like the whole Batch is used to estimate the cost of one round of the method, but it's just that for some methods, (like bayesian optimization) it's not clear how many iterations will be needed.