Refactor Microsimulation.get_weights() to always derive from household_weight

[MaxGhenis]

Summary

Microsimulation.get_weights() in policyengine_core/simulations/microsimulation.py hardcodes f"{entity_key}_weight" to look up weights for each entity. This means every country package must define person_weight, tax_unit_weight, spm_unit_weight, benunit_weight, etc. — even though all of these just delegate to household_weight (the only calibrated weight).

This causes confusion because:

1. Users and AI agents use person_weight, spm_unit_weight, etc. for manual calculations, thinking they're distinct — but they're all just household_weight projected to the entity
2. The existence of multiple weight variables makes it unclear which one to use
3. It's impossible to remove the redundant weight variables without a core change

Current behavior

# microsimulation.py line ~30
weight_variable_name = f"{entity_key}_weight"
weights = self.calculate(weight_variable_name, time_period, use_weights=False)

Every entity type requires a {entity}_weight variable to exist, or Microsimulation crashes.

Proposed behavior

get_weights() should always derive from household_weight, projecting it to whichever entity is requested. Something like:

weights = self.calculate("household_weight", time_period, use_weights=False)
if entity_key != "household":
    weights = self.map_to(weights, "household", entity_key)

This would:

• Eliminate the need for country packages to define per-entity weight variables
• Make it clear that household_weight is the single source of truth
• Reduce confusion for users doing manual weighted calculations

Downstream impact

After this change, country packages could deprecate/remove:

• US: person_weight, tax_unit_weight, spm_unit_weight, family_weight, marital_unit_weight
• UK: person_weight, benunit_weight

The policyengine-api, policyengine.py, and data packages also reference these variables and would need updates.

[baogorek]

I've been creating .h5s in -us-data without explicit weights and I thought there was a mechanism to default to household weights. Maybe that's just in -us.

But even in -us-data, if I put weights in the .h5, the Microsimulatino will use them, and that's been a cause of confusion for me in the past.

The principle is so simple: it's self-weighting. But it is easy to get confused, and I agree that it should at least be harder to change an entity's weights to be different from household weights.

[baogorek]

Right when I thought I was solid on this stuff, Claude shook me:

  - Weight computation: Person weight = household weight; sub-entity weight = (hh_weight / persons_in_hh) × persons_in_entity

I actually sent it this issue, and it responded:

 This settles it. The issue confirms household_weight is the single source of truth, and your own comment flags that explicit weights
   in the H5 override formulas, which has been a source of confusion.                                                                 
                                                                                                                                      
  The cleanest approach: don't store sub-entity weights at all. If person_weight, tax_unit_weight, etc. have formulas in
  policyengine-us that derive from household_weight, then omitting them from the H5 lets those formulas run correctly. Storing them is
   at best redundant, at worst wrong (as with the per-person formula).

Though I realize my own comment above might have influenced it.