[clangd] Correctly identify the next token after the completion point (#69153)

The code was previously using Lexer::findNextToken() which does not
handle being passed the completion point as input.

Fixes https://github.com/clangd/clangd/issues/1785