Saving and looking up metadata key/value pairs seems to work now at
least with some basic tests.

Now _saveIndexKey inserts empty lines when the index isn't long enough.
This is necessary because the page ids are taken from the global page
index, but there is not every page in the metadata key specific index
so e.g. line 10 might be the first entry in the index.

The indexer functions have been converted to a class interface.
Use the Doku_Indexer class to access the indexer with these public methods:
  addPageWords
  addMetaKeys
  deletePage
  tokenizer
  lookup
  lookupKey
  getPages
  histogram
These functions are provided for general use:
  idx_get_version
  idx_get_indexer
  idx_get_stopwords
  idx_addPage
  idx_lookup
  idx_tokenizer
These functions are still available, but are deprecated:
  idx_getIndex
  idx_indexLengths
All other old idx_ functions are unsupported and have been removed.

An external tokenizer inserts extra spaces to mark words in the input text.
The text is sent through STDIN and STDOUT file handles.

A good choice for Chinese and Japanese is MeCab.
http://sourceforge.net/projects/mecab/
With the command line 'mecab -O wakati'

When updating a single line that line was split into an array and in a
loop over that array one entry was removed and afterwards a new one
added. Tests have shown that using a regex for doing that is much faster
which can be easily explained as that regex is very simple to match
while a loop over an array isn't that fast. As that update function is
called for every word in a page the impact of this change is
significant.

This adds a simple boolean variable that tracks if new words have been
added. When editing a page in many cases all words have already been
used somewhere else or just one or two words are new. Until this change
all words indexes read were always written, now only the changed ones
are written. The overhead of the new boolean variable should be low.

In PHP versions newer than 4.3.0 fgets reads a whole line regardless of
its length when no length is given. Thus the loop in _freadline isn't
needed. This increases the speed significantly as _freadline was called
very often.

From my experience with a benchmark of the indexer it is faster to first
join the array of all index entries and then write them back together
instead of writing every single entry. This might increase memory usage,
but I couldn't see a significant increase and this function is also only
used for the small index files, not for the large pagewords index.

p_purge_metadata now updates the metadata cache and the INFO array like the other metadata writing functions

this is mainly for use with plugin:upgrade and thus will be
cherrypicked into the current stable branch as well.

make custom buttons possible with html_btn() without the need of global $lang (more consistent with tpl_pagelink())