Exclude SIPP allocation flags from tip income sum (closes #524)

[MaxGhenis]

Summary

Fixes #524.

sipp.py computed tip income by summing every column containing "TXAMT", which matched both:

• TJB<n>_TXAMT — the per-job tip dollar amounts (the intended source)
• AJB<n>_TXAMT — Census allocation flags (small integers 0/1/2 indicating whether the value was imputed)

Allocation-flag integers were added to dollar totals. The effect is small (the flags range 0–2, so typical inflation per job is <$3/month × 12 = <$36/year) but incorrect.

Fix

# before
df["tip_income"] = (
    df[df.columns[df.columns.str.contains("TXAMT")]].fillna(0).sum(axis=1) * 12
)

# after — match dollar-amount columns only
df["tip_income"] = (
    df[df.columns[df.columns.str.match(r"TJB\d_TXAMT")]].fillna(0).sum(axis=1)
    * 12
)

Implements the fix proposed in #524 verbatim.

Regression tests

New tests/unit/datasets/test_sipp_tip_columns.py:

• test_tip_regex_matches_dollar_amounts_only — the new regex matches TJB1_TXAMT/TJB2_TXAMT but excludes AJB1_TXAMT, AJB2_TXAMT, SOME_TXAMT_OTHER, and TPTOTINC.
• test_tip_sum_excludes_allocation_flags — row-sum check: with dollar amounts [100, 50]/[200, 75] and allocation flags [1, 0]/[2, 1], the new filter yields 150/275 while the old contains("TXAMT") would have yielded 151/278.

Test plan

• 2 new unit tests pass.
• CI passes.