[anqorithm]

Fixes #63

The issue is right about both lines. Checked against Unicode 16:

• ࣔ-࣡ and ࣔ-࣭ overlap, and together they match U+08E2 ARABIC DISPUTED END OF AYAH. That's a format character (Cf), not a mark, so it was deleted along with the harakat.
• 19 combining marks weren't matched: U+0897–089F and U+08CA–08D3, which were added in Unicode 9 to 15. When one of them sits between two letters, the letters stop joining. For example, ب + U+08D0 + ب gave two isolated behs instead of initial and final.

The extended ranges are now ࢗ-࢟, ࣊-࣡ and ࣣ-ࣿ. Every Arabic Mn character in U+0600–08FF is now matched.

Behaviour change to check: U+08E2 is no longer treated as a haraka, so delete_harakat keeps it. That seems right for a Quranic sign, but it's your call. I left U+06E5 and U+06E6 (small waw and small yeh, Lm) as they were.

Testing: added test_006_harakat_ranges.py, which checks joining across the new marks and that U+08E2 is kept. Both tests fail without this change, and the whole suite passes. I tested on Python 3.11, which ships Unicode 14, so U+0897 (added in 15) is covered by the regex but wasn't checked against a Unicode 15 database.

CI follow-up: restrict the credentialed Claude review job to branches in this repository. Fork PRs cannot receive its OAuth secret or write OIDC token; the ordinary test workflow remains enabled for forks. Local validation: 20 passed, 3 optional-dependency skips, plus actionlint. Upstream CI now passes on Python 3.10–3.15; the credentialed review job is skipped for this fork.

[anqorithm]

Fixed the fork-only review failure in 55cc668. The credentialed Claude review job now runs only for same-repository branches; ordinary CI continues to test fork PRs. Upstream CI passes on Python 3.10–3.15, and the Claude job correctly skips for this fork. Locally: 20 tests passed, 3 optional-dependency skips, and actionlint passed.