Handle rotated text

[jbaiter]

Currently we don't include any heuristics for detecting text that is rotated, such as in this example from a Wellcome volume:

Due to this, we stretch the text content to fill the available width, which results in this garbled output (which also takes a long time to render for some reason):

For this case it probable suffices to include a simple heuristic to detect 90-degree rotated text. Since the OCR does not include any indication of the direction of the rotation, it should probably fall back to counter-clockwise due to the reading direction.

Obviously this should only apply to texts in western LTR scripts, Asian scripts that are written top-down would obviously be rendered badly with this approach.

I don't think we can "fix" this without help from the actual OCR markup. Even in the western script case, a line might be rotated 90 or 270 degrees, there's no way of telling from this markup, since it includes neither baseline nor rotation information:

<TextBlock ID="P9_TB00006" HEIGHT="316" WIDTH="57" HPOS="649" VPOS="651" STYLEREFS="TXT_43 PAR_CENTER">
  <TextLine ID="P9_TL00007" HEIGHT="316" WIDTH="31" HPOS="649" VPOS="651">
    <String ID="P9_ST00034" CONTENT="Vestry.—33" HEIGHT="167" WIDTH="31" HPOS="649" VPOS="800" WC="0.93" CC="0700000000"/>
    <SP ID="P9_SP00028" WIDTH="0" HPOS="674" VPOS="763"/>
    <String ID="P9_ST00035" CONTENT="Meetings." HEIGHT="133" WIDTH="30" HPOS="650" VPOS="651" WC="1" CC="000000000"/>
  </TextLine>
  <TextLine ID="P9_TL00008" HEIGHT="287" WIDTH="25" HPOS="681" VPOS="666">
    <String ID="P9_ST00036" CONTENT="No." HEIGHT="49" WIDTH="23" HPOS="681" VPOS="904" WC="1" CC="000"/>
    <SP ID="P9_SP00029" WIDTH="0" HPOS="706" VPOS="874"/>
    <String ID="P9_ST00037" CONTENT="of" HEIGHT="27" WIDTH="15" HPOS="690" VPOS="861" WC="1" CC="00"/>
    <SP ID="P9_SP00030" WIDTH="0" HPOS="706" VPOS="828"/>
    <String ID="P9_ST00038" CONTENT="Attendances." HEIGHT="184" WIDTH="22" HPOS="682" VPOS="666" WC="1" CC="000000000000"/>
  </TextLine>
</TextBlock>