This should probably be untangled into 3 separate patches, but
looking for feedback first on whether we want to do this at all.
a) store include-insertion information for main-file symbols that are
not eligible for code completion indexing.
b) consider an #ifndef guard that's dangling at the end of the preamble
to be a well-formed header guard (otherwise we never will)
c) load HeaderFileInfo for the preamble main-file even if the file-size
mismatches
headerSymbols still uses Filename not HeaderFilename of the TU.