Add clean_name utility and fallback_func support to lookup

[jcarbaugh]

Summary

Adds a clean_name text-normalization helper and a pluggable fallback_func mechanism to us.states.lookup, plus a cache-hit short-circuit.

Closes #59

What changed

clean_name(text) — strips an incoming string down to a bare state name: lowercases, replaces punctuation and underscores with spaces, drops the filler words the, commonwealth, state, and of, and recombines the remaining tokens into a single space-separated string. For example, " The state OF idaho " becomes "idaho". It is a standalone helper — lookup does not call it automatically, so callers opt in with lookup(clean_name(raw)).

fallback_func argument on lookup — lookup accepts an optional fallback_func, defaulting to None. When the normal FIPS/abbreviation/metaphone matching finds nothing, fallback_func(val) is called with the original, unmodified lookup value and may return a State or None. The fallback decides for itself what to match against. Existing behavior is unchanged when no fallback is passed.

startswith_fallback(val) — a ready-made fallback_func that matches val against the start of each state's or territory's name, case-insensitively, returning the first match (or None, including for empty input).

Approach notes

• lookup mutates its val argument in place (upper-casing abbreviations, metaphone-encoding names). The original value is captured up front and passed to fallback_func, so fallbacks receive the human-readable string rather than a metaphone code.
• lookup now returns immediately on a cache hit instead of scanning the full state list afterward — a pre-existing inefficiency.
• Fallback hits are cached under a fallback-specific key, so a cached fallback result is never returned to a lookup that passes a different fallback or none at all.

Testing

Adds tests for clean_name, startswith_fallback, fallback resolution and caching semantics, and the cache-hit short-circuit. uv run pytest . passes (30 passed, 1 skipped — a pre-existing network test); ruff check and ruff format --check are clean.