tjDecompressHeader3: allow loading an abbreviated table specification datastream

[amyspark]

This allows libjpeg-turbo to prime the decoder with an
abbreviated table specification datastream from a TIFFTAG_JPEGTABLES
field, a capability that was until now only available with raw libjpeg.

Fixes #604

[dcommander]

Sorry for the delay. A few further questions:

1. Would it be OK to just return 0 for success, regardless of whether a full image or an abbreviated (tables-only) datastream is passed to the function? My concern is that some existing programs may lazily check the return value like so: if (tjDecompressHeader3(...)) { throw an error; }.
2. How would this new API be used, in practice? Would a program call tjDecompressHeader3() once with a tables-only datastream and then again with the actual image?
3. How are tables-only datastreams generated? I confess that I am not familiar with that aspect of the libjpeg API.

[amyspark]

1. That's OK with me.
2. Indeed, that's how libtiff handles it at present. In my own case:

#ifdef HAVE_JPEG_TURBO
    uint32_t hasSplitTables = 0;
    uint8_t *tables = nullptr;
    uint32_t sz = 0;
    QVector<unsigned char> jpegBuf;

    auto handle = [&]() -> std::unique_ptr<void, decltype(&tjDestroy)> {
        if (planarconfig == PLANARCONFIG_CONTIG
            && color_type == PHOTOMETRIC_YCBCR
            && compression == COMPRESSION_JPEG) {
            return {tjInitDecompress(), &tjDestroy};
        } else {
            return {nullptr, &tjDestroy};
        }
    }();

    if (color_type == PHOTOMETRIC_YCBCR && compression == COMPRESSION_JPEG
        && hsubsampling != 1 && vsubsampling != 1) {
        dbgFile << "Setting up libjpeg-turbo for handling subsampled JPEG...";
        if (!TIFFGetFieldDefaulted(image,
                                   TIFFTAG_JPEGTABLESMODE,
                                   &hasSplitTables)) {
            errFile << "Error when detecting the JPEG coefficient "
                       "table mode";
            return ImportExportCodes::FileFormatIncorrect;
        }
        if (hasSplitTables) {
            if (!TIFFGetField(image, TIFFTAG_JPEGTABLES, &sz, &tables)) {
                errFile << "Unable to retrieve the JPEG abbreviated datastream";
                return ImportExportCodes::FileFormatIncorrect;
            }
        }

        {
            int width = 0;
            int height = 0;

            if (hasSplitTables
                && tjDecompressHeader(handle.get(), tables, sz, &width, &height)
                    != 1) {
                errFile << tjGetErrorStr2(handle.get());
                m_doc->setErrorMessage(
                    i18nc("TIFF errors",
                          "This TIFF file is compressed with JPEG, but "
                          "libjpeg-turbo could not load its coefficient "
                          "quantization and/or Huffman coding tables. "
                          "Please upgrade your version of libjpeg-turbo "
                          "and try again."));
                return ImportExportCodes::FileFormatIncorrect;
            }
        }
    }
#endif

3. That's done by libtiff under the hood, it doesn't affect usage of YCbCr. If I understand the code correctly, on writing, it commands libjpeg to not output coefficient tables via jpeg_suppress_tables, then it redirects the writing of the tables to a separate buffer (TIFFjpeg_tables_dest) before calling jpeg_write_tables.

[dcommander]

Implemented and pushed. Please test.