fix/handle text_special and text_join rule causing incorrect position and size with escaped links

[stanley2058]

Cause

• Escape characters like \ go through rules_core/text_join.mjs and are converted from text_special to text and merged before other rules run. Causing later rules to lose the correct position info.

Fixes

• Update rules_core/text_join.mjs to preserve position and size when merging adjacent text tokens.
• Update token's size when creating a text_special token in rules_inline/escape.mjs to preserve the correct size before stripping the escape token.

Others

• Also fix the regression for introducing the ZWSP character (which causes tests to break and missing trailing \n in some cases).

GPT-5 saved it 🙏