[stantheman0128]

Closes #536.

Summary

TSPhrases.txt still forced a few cover-sense 覆 and mat/disorder-sense 藉 phrases to 复 / 借. That disagrees with the simplified Chinese standard (same basis as #1071 for 射覆):

• 覆 meaning cover / overturn stays 覆
• 藉 meaning mat / disorder / classical non-borrow senses stays 藉
• Reply/repeat 覆 (答覆 / 反覆 / 回覆 / 批覆) still goes to 复
• Borrow/pretext 藉 (藉口 / 憑藉 / 藉端生事 / 狐藉虎威) still goes to 借

Change

data/dictionary/TSPhrases.txt (identity keep):

名覆金甌 / 覆墓 / 覆帳
茵藉 / 素藉 / 藉槁 / 藉甚 / 藉草枕塊 / 躪藉 / 醞藉 / 凌藉 / 滑藉 / 藉卉

data/dictionary/STPhrases.txt: remove the matching reverse entries that taught the incorrect simplified forms (same pattern as #1071).

Left alone on purpose (policy / out of consensus scope): reply-sense 覆 phrases, borrow-sense 藉 phrases, and ambiguous items such as 顧藉 / 覆按.

AI assistance

This PR was prepared with AI assistance (Cursor / Grok). Stan Shih (stantheman0128) reviewed the dictionary choices against the simplified-character rules and the harness output before opening the PR.

Test plan

• New cases BYVoid_OpenCC_Issue_536_* in test/testcases/testcases.json
• Controls: reply-sense 覆 and borrow-sense 藉 still convert
• Local verification with OpenCC 1.4.1 wheel + branch text dictionaries (see Evidence)
• CI Bazel dictionary tests (authoritative)

Evidence

Before (packaged wheel dictionaries):

t2s  名覆金甌 -> 名复金瓯
t2s  覆墓 -> 复墓
t2s  覆帳 -> 复帐
t2s  茵藉 -> 茵借
t2s  藉甚 -> 借甚
t2s  藉草枕塊 -> 借草枕块
t2s  醞藉 -> 酝借

After this branch (config JSON ocd2 refs rewritten to text, driving data/dictionary/*.txt):

t2s  名覆金甌 -> 名覆金瓯
t2s  覆墓 -> 覆墓
t2s  覆帳 -> 覆帐
t2s  射覆 -> 射覆
t2s  答覆 / 反覆 -> 答复 / 反复
t2s  茵藉 / 藉甚 / 藉草枕塊 / 醞藉 -> 茵藉 / 藉甚 / 藉草枕块 / 酝藉
t2s  藉口 / 憑藉 / 狐藉虎威 -> 借口 / 凭借 / 狐借虎威

Harness summary (OpenCC 1.4.1 wheel, text-dict harness):

branch dicts + new #536 cases: 4 passed, 0 failed
existing t2s expectations: 52 pass; 3 pre-existing text-harness gaps (rare-char stubs), unchanged
sort.py idempotent on TSPhrases/STPhrases: no further diff