Add tracing regression coverage for automatic chooser-column pruning

[jpn--]

Problem

PR #1094 makes automatic unused-column pruning run while tracing is enabled. This closes the memory gap left by #833, where pruning was intentionally bypassed for traced runs, but it also means tracing now depends on preserving the correct identity columns.

The fix in 0f5234e adds util.traceable_id_columns() and protects:

• household_id
• person_id
• proto_household_id
• proto_person_id

across the affected simple- and interaction-simulation paths.

The tests added in #1094 only partially cover this behavior. They inspect the pruned column list with tracing mocked; the interaction_sample test also mocks interaction_trace_rows(). Consequently, they do not verify that real trace targets are selected after pruning, that proto-table tracing still works, or that all affected simulation paths preserve the required identifiers.

The integration test added by #1021 verifies consistent trace filename hashes across multiprocessing, but does not validate the rows or identifiers contained in trace output.

This follows up on:

• Deprecate SIMULATE_CHOOSER_COLUMNS and LOGSUM_CHOOSER_COLUMNS settings #1094 and the review at Deprecate SIMULATE_CHOOSER_COLUMNS and LOGSUM_CHOOSER_COLUMNS settings #1094 (review)
• Automatically drop unneeded columns in choosers table #833, which introduced automatic chooser pruning
• Tracing fix for HH ID that exists in both synthetic and proto populations #901, which added proto-household/proto-person tracing
• Fix trace IDs so they have the same hash #1021, which added the existing trace-ID integration test

Proposed coverage

Add focused regression tests that exercise tracing without mocking the trace-row selection logic.

1. Add table-driven tests for Tracing.interaction_trace_rows() covering:

   • person_id and household_id as an index or chooser column
   • proto_person_id and proto_household_id, including proto-table precedence
   • fixed-size interaction alternatives (sample_size set)
   • sparse or variable-size alternatives (sample_size=None)
   • the expected trace-row mask and returned trace IDs

2. Extend the pruning tests for the paths changed by Deprecate SIMULATE_CHOOSER_COLUMNS and LOGSUM_CHOOSER_COLUMNS settings #1094:

   • simple simulation and logsum calculation
   • interaction_sample
   • interaction_simulate
   • interaction_sample_simulate

3. Verify that:

   • spec-used columns remain available
   • unrelated chooser columns are pruned before constructing/evaluating the interaction dataframe
   • every identifier required for trace slicing survives pruning
   • the configured household/person is actually selected in the resulting trace rows
   • the full, unpruned chooser and alternative inputs are still written before pruning, as intended by Deprecate SIMULATE_CHOOSER_COLUMNS and LOGSUM_CHOOSER_COLUMNS settings #1094

4. Strengthen the lightweight integration coverage, either by extending test/trace_id or adding a small equivalent test, to inspect trace contents rather than only filename hashes.

Acceptance criteria

• Tests exercise real trace-row selection rather than substituting a mocked result.
• Regular and proto household/person identifiers are covered.
• Removing trace-ID protection from any affected simulation path causes a regression test to fail.
• At least one test validates emitted trace data or captured trace_df input, not only the intermediate column list.
• The tests use small synthetic fixtures and remain suitable for the normal core test suite.