[frankslin]

Fills a gap where TWPhrases/HKPhrases had no coverage for well-known film industry figures, causing s2twp/s2hkp to leave their names in generic Traditional form instead of the region-specific transliteration readers actually expect.

Detailed Changes:

• Selection criteria:
  • Names were sourced from Chinese Wikipedia's community-maintained NoteTA regional-name tables (cross-checked via Wikidata labels where NoteTA parsing was ambiguous), not from memory, to avoid introducing fabricated transliterations.
  • Candidates came from two Wikipedia categories: Academy Award for Best Director winners, and Academy Award for Best Picture winning producers (~16 people overlap between the two sets).
  • Entries were only added where the regional form is a genuine transliteration difference (e.g. 卡梅隆 vs 卡麥隆 vs 金馬倫), not merely a simplified/traditional script difference (e.g. Woody Allen's TW name, George Lucas's TW name were excluded since they are identical to the Mainland form once script is normalized).
  • Each dictionary key is the generic Traditional form (the true s2t/STCharacters output), matching the existing file convention used throughout TWPhrases.txt/HKPhrases.txt, not the raw Simplified input.
• Full name vs. surname-alone:
  • Default to full name (with · separator) for every entry, mirroring the project's one existing precedent (沃恩·威廉斯 → 佛漢·威廉斯), since a bare surname risks colliding with unrelated people who share a common Western surname (Cameron, Jackson, Scott, etc. all appear on other well-known figures with potentially different accepted translations).
  • Added exactly one surname-only entry, 诺兰 → 諾蘭 (TW) / 路蘭 (HK), because it passed a stricter bar: the surname is dominant in casual Chinese usage for this one person, and cross-checking his brother Jonathan Nolan's Wikipedia entry confirms the same surname transliteration is used consistently across different bearers of the name, i.e. it is a stable, person-independent convention rather than an entity-specific choice.
  • All other surnames were judged too collision-prone standalone (e.g. 卡梅隆/Cameron also names a UK PM, 傑克森/Jackson also names Michael Jackson) and were kept as full-name-only.
• Regression test cases: added testcases.json entries that feed Simplified input through s2t/s2tw/s2hk/s2twp/s2hkp together in a single case per name group, plus tw2sp/hk2sp round-trips. This replaces an earlier, narrower set of cases that only checked s2twp/s2hkp and missed a real bug: keying TWPhrases/HKPhrases by raw Simplified text (instead of the generic Traditional form) caused STPhrases_GeneratedFromRegionalPhrases to emit a Simplified-to-itself identity mapping that short-circuited plain s2t/s2tw/s2hk for these names. Testing all five modes together in one case now catches that class of regression immediately.