xlsx: fix ignoring rich text <r> after initial plain <t>

[aquasync]

Currently read_string_with_bufs just ignores all subsequent events until closing tag if hitting a plain text node first. This causes it to ignore subsequent text in examples like this:

  <si>
    <t>tval</t>
    <r>
      <t>rval1</t>
    </r>
    <r>
      <t>rval2</t>
    </r>
  </si>

I guess this was for speed reasons? I think in the typical case the next event would be the closing tag anyway, so shouldn't be much different, and I don't think this introduces any additional copies. It does make rich_buffer a bit of a misnomer though, so maybe should be changed (though text_buf means something different in this function).