[LeeiFrankJaw]

In Taiwan standard the beam sense of 樑 is written as 梁, so it collapses
with the surname/place-name 梁. Character-level TWVariantsRev cannot tell
the two apart on the way back, which left tw2t (and the tw2* segmentation
chain) unable to restore 樑 for beam phrases — 橋梁 round-tripped to 橋樑
via s2t but stayed 橋梁 via tw2t, an inconsistency with STPhrases.

This adds the missing phrase-level reverse mappings so that whole beam
phrases are recovered from context, while genuine 梁 words (surnames,
place names) are preserved.

Detailed Changes:

• TWVariantsRevPhrases.txt:
  • Add 41 beam-sense entries mirroring the 梁 -> 樑 phrases in STPhrases
    (e.g. 上梁文 -> 上樑文, 國家棟梁 -> 國家棟樑, 橫打鼻梁兒 -> 橫打鼻樑兒),
    so tw2t converts the Taiwan 梁 form back to the OpenCC-standard 樑.
  • Keep self-mapping entries where 梁 must be preserved (白鶴梁, 梁木,
    梁龍, ...) so proper nouns are not over-converted.
• test/testcases/testcases.json:
  • Add a tw2t case covering all 41 new entries, guarding the beam-vs-name
    distinction against regressions.