In Reproduce in Python code, when simulating over a state, properly update dataset

[anth-volk]

For a select subset of reforms that we run over states (e.g., https://policyengine.org/us/policy?focus=policyOutput.policyBreakdown&reform=66051&region=ny&timePeriod=2025&baseline=2&household=49726), the API processes the reforms by swapping the default cps_2024 dataset for the pooled_3_year_cps_2023 dataset and filtering the dataset down by state. At the moment, the Reproduce in Python code does not reflect that, causing the app to output total values that differ from those of the -us package.

This should be pretty simple - we merely need to update the Microsimulation objects to take dataset='pooled_3_year_cps_2023' as an argument. However, we'll need to determine if we also want to add code to pre-filter these datasets, since not doing so causes an individual run to take just under 5 minutes.