Fix enum encoding for UTF-8 byte strings

[anth-volk]

Fixes #487

Context

This came up while comparing a staged PolicyEngine US data publication. The staged US.h5 file stored county/2024 as an HDF5 fixed-width byte-string array, and 15 rows contained the valid enum member name DOÑA_ANA_COUNTY_NM as UTF-8 bytes:

b"DO\xc3\x91A_ANA_COUNTY_NM"

policyengine-us already has DOÑA_ANA_COUNTY_NM in its county enum, and normal situation input with the Unicode string works. The failure was specific to dataset-backed simulation loading: h5py returns HDF5 string data as bytes, then Core's enum encoding path used array.astype(str), which attempts ASCII decoding and raises before Microsimulation(dataset=...) can finish constructing.

The intended conversion is:

HDF5 UTF-8 bytes -> Unicode enum member name -> EnumArray integer code

Summary

• Decode fixed-width byte-string arrays as UTF-8 before enum-name lookup.
• Add a regression test for HDF5-style UTF-8 byte strings with a non-ASCII enum member name.
• Document in the test that the ñ character is intentional because it exercises the non-ASCII byte-string decode failure.
• Add provider-neutral AI guidance for tests, documentation review, PR creation, and changelog fragments so agents check and add towncrier fragments before opening PRs.
• Add towncrier changelog fragments under changelog.d/.

Testing

• uv run pytest tests/core/enums/test_enum.py
• uv run ruff format policyengine_core/enums/enum.py tests/core/enums/test_enum.py
• uv run ruff check policyengine_core/enums/enum.py tests/core/enums/test_enum.py
• uv run --no-sync --with towncrier towncrier build --draft --version 0.0.0