fix(omml2latex): 修复多字母标识符被拆开的问题

OMML2MML 会把 Word 里连写的多字母标识符（线段名 BC、三角形 △ABC、
角度 45° 等）拆成相邻的 <mi>/<mn> 元素，原转换器用空格 join 导致
'BC' 被渲染成 'B C'（B×C），'45°' 变成 '4 5°'。

三处修复：
- mrow_visitor: 连续单字母 <mi> 拼接
- legacy_adapter: <math> 根节点多子元素时连续单字母 <mi> 拼接
- math_strategy: Unicode+字母混合内容按'LaTeX命令后跟字母才加空格'规则 join

回归验证：test_mixed.docx 130 个 OMML 公式，16 个发生变化且全部为
正确改善（线段名/角度/三角形/隐式乘法拼接），零误伤。

scripts/omml_mathml2latex/compat/legacy_adapter.py

@@ -326,6 +326,19 @@ class LegacyMathMLToTeXConverter:
                         # Found fragmented function! Add it as \funcname
                         parts.append(f"\\{function_name}")
                         i += consumed  # Skip the consumed elements
+                    elif self._is_plain_single_letter_mi(children[i]):
+                        # Consecutive single-letter identifiers are joined without spaces:
+                        # Word often stores a multi-letter identifier (line segment "BC")
+                        # as a single <m:t> that OMML2MML splits into adjacent <mi> runs.
+                        # Re-merge them so "BC" stays "BC" instead of "B C" (B×C).
+                        merged = self._convert_node(children[i])
+                        j = i + 1
+                        while j < len(children) and self._is_plain_single_letter_mi(children[j]):
+                            merged += self._convert_node(children[j])
+                            j += 1
+                        if merged.strip() not in ['', '\\']:
+                            parts.append(merged)
+                        i = j
                     else:
                         # Normal element - convert and add
                         part = self._convert_node(children[i])
@@ -804,6 +817,24 @@ class LegacyMathMLToTeXConverter:
         chemical_indicators = mathrm_count + number_count
         return chemical_indicators > len(parts) / 2
 
+    def _is_plain_single_letter_mi(self, node: ET.Element) -> bool:
+        """Return True if node is a single plain-letter <mi> identifier.
+
+        Used to re-merge multi-letter identifiers (e.g. line segment "BC") that
+        Word/OMML2MML split into adjacent <mi> runs. Only plain italic identifiers
+        are merged; elements carrying an explicit mathvariant (e.g. normal for
+        chemical formulas) are left untouched.
+        """
+        if not isinstance(getattr(node, 'tag', None), str):
+            return False
+        if self._get_local_name(node.tag) != 'mi':
+            return False
+        if node.get('mathvariant') is not None:
+            return False
+        from ..utils import text_content
+        content = text_content(node).strip()
+        return len(content) == 1 and content.isalpha()
+
     def _get_local_name(self, tag: str) -> str:
         """
         Extract local element name from tag (strip namespace).

scripts/omml_mathml2latex/strategies/math_strategy.py

@@ -141,7 +141,19 @@ class MathStrategy(Strategy):
                         converted_parts.append(' ')
                     else:
                         converted_parts.append(char)
-                return ' '.join(converted_parts)
+                # Join with intelligent spacing: a LaTeX command followed by a
+                # letter/digit needs a space (otherwise "\triangleA" becomes an
+                # undefined command), but plain letters stay adjacent so that a
+                # multi-letter identifier like "△ABC" becomes "\triangle ABC",
+                # not "\triangle A B C".
+                result_parts = []
+                for idx, part in enumerate(converted_parts):
+                    result_parts.append(part)
+                    if idx < len(converted_parts) - 1:
+                        nxt = converted_parts[idx + 1]
+                        if part.startswith('\\') and nxt and (nxt[0].isalpha() or nxt[0].isdigit()):
+                            result_parts.append(' ')
+                return ''.join(result_parts)
 
         # USER REQUIREMENT: "use the mathvariant to determine whether it's mathrm"
         # REMOVED: Chemistry detection heuristics (lines 139-162)

scripts/omml_mathml2latex/visitors/mrow_visitor.py

@@ -1633,6 +1633,18 @@ class MrowVisitor(Visitor):
                 # Found a fragmented function - add it to parts
                 parts.append(function_name)
                 i += consumed_count
+            elif self._is_plain_single_letter_mi(children[i]):
+                # Consecutive single-letter identifiers are joined without spaces:
+                # Word often splits a multi-letter identifier (e.g. line segment "BC")
+                # into several runs, which OMML2MML turns into adjacent <mi> elements.
+                # Re-merge them so "BC" stays "BC" instead of "B C" (which renders as B×C).
+                merged = self._convert_child(children[i])
+                j = i + 1
+                while j < len(children) and self._is_plain_single_letter_mi(children[j]):
+                    merged += self._convert_child(children[j])
+                    j += 1
+                parts.append(merged)
+                i = j
             else:
                 # No fragmented function - convert child normally
                 part = self._convert_child(children[i])
@@ -1642,6 +1654,21 @@ class MrowVisitor(Visitor):
 
         return ' '.join(parts).strip()
 
+    def _is_plain_single_letter_mi(self, node: ET.Element) -> bool:
+        """Return True if node is a single plain-letter <mi> identifier.
+
+        Used to re-merge multi-letter identifiers (e.g. line segment "BC") that
+        Word split into adjacent <mi> runs. Only plain italic identifiers are
+        merged; elements carrying an explicit mathvariant (e.g. normal for
+        chemical formulas, double-struck, etc.) are left untouched.
+        """
+        if _local(node.tag) != 'mi':
+            return False
+        if node.get('mathvariant') is not None:
+            return False
+        content = _text_content(node).strip()
+        return len(content) == 1 and content.isalpha()
+
     def _convert_children_no_space(self, node: ET.Element) -> str:
         """Convert all children and join without spaces."""
         parts = []