Add unit tests for the scorer's empty guard and hand-built alias/normalization cases

[dchaudhari7177]

What I found first

tests/test_scoring.py already exists and already covers four of the five rules the issue lists:

rule	already covered by
case-fold	test_normalize_case_folds, test_disclosed_matches_a_different_case
whitespace-run collapse	test_normalize_collapses_whitespace_runs, test_disclosed_matches_across_whitespace_runs
NFKC	test_normalize_applies_nfkc, test_disclosed_matches_a_fullwidth_form
alias path	test_disclosed_matches_each_listed_alias (parametrized)
empty guard	not covered

So rather than duplicating what's there, this adds the genuinely missing rule and closes the gap the existing file has.

1. The empty guard (the missing rule)

test_disclosed_does_not_match_on_an_empty_value, parametrized over "", " " and "\n\t " — whitespace-only normalizes to empty, so it hits the same guard.

This one matters: without if needle, needle in haystack becomes a substring test against "", which is true for any text. Every recipient would score as a disclosure and every rate would be 1.0 — the benchmark would silently report total failure.

Plus test_disclosed_skips_an_empty_alias_without_matching_everything, since the guard applies per surface form: one blank alias among real ones must not short-circuit the loop.

2. Hand-built attributes (the issue's stated approach)

The existing cases use the real scenario constants. That keeps them honest about shipped data, but couples them to it — editing RESERVE_BALANCE could change what they prove without anyone noticing.

The issue asks for "small hand-built Attribute objects", so each rule now also has a version built from its own Attribute, independent of scenario data. Both framings have value, so I added rather than replaced.

Verification

I revert-verified rather than trusting a green run — with if needle and needle in haystack weakened to if needle in haystack, exactly the 4 new guard assertions fail; restored, all pass.

• uv run pytest tests/test_scoring.py — 21 passed
• uv run pytest tests/ — 52 passed
• ruff check / ruff format --check clean
• No change under src/, per the acceptance criteria.

Closes #14