Add Arabic language pack (ara) — proclitic tokenization + orthographic normalization

[bsesic]

Summary

Adds the Arabic language pack (ara) — the first non-Hebrew language pack, exercising and validating the LanguagePack abstraction. Closes #16.

Rule-based and dependency-light (stdlib + existing rapidfuzz only), mirroring the structure of src/tracealign/lang/hebrew/.

What's included

• Tokenization (tokenize.py): conservative proclitic splitting — splits the definite article ال, single-letter proclitics (و/ف/ب/ك) only before the article, and the لل (li- + article) contraction. Deliberately does not amputate word-initial radicals (وزير, باب, كتاب). Spans are contiguous (no separator character).
• Normalization (normalize.py): strip_tashkil (NFC, remove combining marks + tatweel) into text; orthographic skeleton folding (alif variants → ا, ة → ه, ى → ي, hamza seats, bare hamza dropped). Diplomatic form preserved in Token.raw.
• Scoring (scoring.py): four tiers — EXACT → DIACRITICS_STRIPPED → ORTHOGRAPHIC_VARIANT (skeleton) → ORTHOGRAPHIC (fuzzy).
• Reason enum: two new script-neutral values — DIACRITICS_STRIPPED, ORTHOGRAPHIC_VARIANT.
• Pack + registration: ArabicLanguagePack (code="ara", aliases=("arabic",), version="ara-0.1.0", empty Lexica()), wired via the side-effect-import + _BUILTIN_PACK_MODULES pattern.

Design notes

• Conservative splitting chosen over maximal: over-splitting damages alignment; under-splitting is recoverable by the fuzzy tier. This also means no guard lexicon / data files are needed.
• Full design rationale: docs/superpowers/specs/2026-06-24-arabic-language-pack-design.md.

Testing

• 227 tests passing; new Arabic package has 33 tests (tokenize incl. negative + لل + كال cases, normalize, scoring tiers, registry, end-to-end align with language_pack_version in params).
• flake8 src/ tests/ clean.
• Verified green on Python 3.10 and 3.12.

Follow-up (non-blocking, from review)

• Cross-pack consistency of the fuzzy-tier details["layer"] label (Arabic "fuzzy" vs Hebrew "orthographic") — left as a deliberate design question.