fix and enhance ODF::lpOD

Changes for 6.000

• "Character mode" is now the default and the old ':chars' import tag is deprectated. Now you only need to use ODF::lpOD; use ODF::lpOD_Helper; A new import tag ':bytes' preserves the old ODF::lpOD behavior where method args and results must be encoded binary.
• Introduced methods Hget_text, Hoffset_into_vtext, Hinsert_element, Hsplit_element_at, Hdescendants, Hnext_elt, Hnormalize
• A new concept is a "prune condition" when searching or collecting virtual text, to treat specified tags as opaque (i.e. their children are not visited). This can be used to skip over nested paragraphs such as may exist inside Frames or Tables.
• Results from multi-segment substitutions are now comprehensively "normalized", eliminating consecutive or zero-length text:s segments and ensuring that consecutive spaces are not stored in #PCDATA nodes.
• New style spans are "hoisted" to the top level in a paragraph to avoid creating nested spans (existing spans are cloned and partitioned if necessary around hoisted new spans). NOTE: Please let me know if you *do* want nested spans -- it should be easy to make hoisting an option.