made fixes for really broken but readable files.

[mlaukala]

Can now open and parse files with an incorrect startxref and incorrect stream lengths. Note that adobe acrobat x 10.0.0 will open these files and prompt for save when closed.

When the startxref is incorrect, it looks for the 'trailer' symbol and uses that trailer. Otherwise the xref table is not rebuilt. When the trailer is found, it then parses through the entire file and records the location of each object and places a new PdfReference inside of the PdfCrossReferenceTable.

Will also attempt to correct invalid stream lengths. After the stream length is pulled from the object, we first check for an incorrect 'endstream' symbol. If the 'endstream' symbol is not present where expected, we then look for the next valid 'endstream' symbol after the 'startstream' symbol. We use the 'endstream' symbol index and set the length of the stream.

Note 1: No implementation for a pdf file with a compressed trailer object yet.
Note 2: Not tested with versioned files and will still probably fail.
Note 3: On invalid stream length, should probably check 1k chunks of data for 'endstream'. Currently checks within the invalid length and if not found, loads the rest of the file and checks again.

[mlaukala]

ScanNextToken() will usually crash if it finds itself inside of a stream due to unexpected characters. Currently, my added function IsValidXref() uses ScanNextToken() to determine if the next token is an 'xref' token. This is used to determine if a fix needs to be made. However, if startxref is wrong, the _lexer.Position could be inside of a stream which will more than likely cause an exception.