[clangd] Fix a crash in semantic highlighting.

[hokein]

Summary

We encounter a few internal reports that we're dereference a nullptr.
Unfortunately, no small reproduce testcase for this crash yet, but it makes the
clangd more robost on broken code.

[kadircet]

clang-tools-extra/clangd/SemanticHighlighting.cpp

525–526

can you add a comment here "We might have offsets in the main file that don't correspond to any spelled tokens." i think we should fix that in token buffer spelled token mappings going forward. we're relying on other places around spelledTokenAt returning a valid Token whenever we pass it a main file source location.

[hokein]

add a comment.

clang-tools-extra/clangd/SemanticHighlighting.cpp

public:
 
   const HeuristicResolver *getResolver() const { return Resolver; }
 
 private:
   std::optional<Range> getRangeForSourceLocation(SourceLocation Loc) {
     Loc = getHighlightableSpellingToken(Loc, SourceMgr);
     if (Loc.isInvalid())
       return std::nullopt;
-
+    // We might have offsets in the main file that don't correspond to any
+    // spelled tokens.
     const auto *Tok = TB.spelledTokenAt(Loc);
-    assert(Tok);
-
+    if (!Tok)
+      return std::nullopt;
     return halfOpenToRange(SourceMgr,
                            Tok->range(SourceMgr).toCharRange(SourceMgr));
   }
 
   const syntax::TokenBuffer &TB;
   const SourceManager &SourceMgr;
   const LangOptions &LangOpts;
   std::vector<HighlightingToken> Tokens;