Refine reconstruction of xref tables and trailers

[m-holger]

When xref table recovery is triggered by failure to find a valid trailer, accept the last valid trailer rather than the first valid trailer.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.33%. Comparing base (973edb4) to head (5de67d4).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1196      +/-   ##
==========================================
- Coverage   93.34%   93.33%   -0.01%     
==========================================
  Files         299      299              
  Lines       32997    33000       +3     
  Branches     4037     4038       +1     
==========================================
+ Hits        30801    30802       +1     
- Misses       2196     2198       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[m-holger]

xref table creation and reconstruction raises a number of questions, including:

1. QPDF objects without a root object are generally unusable and will eventually throw a damagedPDF exception. Would it make sense to check for the root object early?

2. Should a trailer without /Root key be accepted as a valid trailer (which would necessarily end in a damagedPDF exception)? It would seem to make sense to treat such trailers as damaged and try to find a better candidate.

3. qpdf generates a warning if the highest object number in the xref table does not match the trailer /Size key. The spec states

    Any object in a cross-reference section whose number is greater than this
    value shall be ignored and defined to be missing by a conforming reader.
   

   Should qpdf remove objects with ids >= /Size from the xref table?