This patch adds an option to choose how filenames are encoded
when saved to the file system. You can choose between urlencoding
(url), the new SafeFn method (safe) and storing real UTF-8 (utf-8).

This class tries to satisfy the following requirements:

  * all ASCII alphanumeric chars in the input should stay the same
    ASCII alphanumeric chars in the output
  * the resulting string should be as short as possible
  * the operation needs to be reversable without any data loss
  * the resulting ASCII string should be case insensitive
  * there should be no restriction on the input length
  * the whole UTF-8 range should be allowed

using it creates a way to store UTF-8 in filenames even if the
underlying filesystem does not support UTF-8. It is also pretty
robust when files are moved between various filesystems and it
creates shorter filenames than the currently used urlencoding.

The replacements (@...@) were partially translated and had spaces in
between. Please don't translate the text between @...@.

This introduces a new function called tpl_flush() which should be used
by template authors in favor of the PHP flush() method.

Plugins may return false in isSingleton to let plugin_load return a new
instance every time it is called.
Renderer plugins are not loaded with $new set to true, but instead specify
themself that they are not singletons. This behaviour allows the odt renderer
to keep working (see #1598).

50e988b1 accidentally stopped act_dispatch from calling act_edit for locked
pages, thus showing a generic »page not writable« message instead of a
page lock message.

explode converts $string (false) to string ("") and returns it if $delimiter is
not found, therefor looking for the key "" in metadata array.

Each searches on the wiki use this function. Scanning the index directory eachtime is time consuming with a constant series of disk access.
Switching a normal search to use file_exists 1 or more times, and not readdir all the directory.
Switching a wildcard search to use a lengths.idx file containing all the word lengths used in the wiki, file generated if a new configuration parameter $conf[readdircache] is not 0 and fixed to a time in second. Creation of a new function idx_listIndexLengths to do this part.

This commit adresses two issues:
  * rawWikiSlices used to drop the first byte of the suffix (usually a newline,
    but custom section edits may not be bound to newlines)
  * con used to insert way too much newlines: the newline that got
    dropped by the bug in rawWikiSlices and one additional newline in prefix
    and text each if they are not terminated by a newline. Now con only inserts
    newlines if there are absolutely no newlines between prefix and text and
    text and suffix.

It seems that the 0x0B Vertical Tab character breaks the PHP
XML parser. This workaround replaces the char with a space
before parsing. Not ideal but good enough for now.

The hash option did not hash the correct content for the
found media files. This causes troubles in the sync plugin
http://github.com/splitbrain/dokuwiki-plugin-sync/issues#issue/1

When the page is not editable or the section info is out of date, a custom
editor should not be used.

The constants are required by the class constructor, which effectively
means before the autoloader is triggered.  This change fixes that issue.

The new COMMON_PAGE_FROMTEMPLATE is triggered by pageTemplate AFTER the template
has been read but before performing the template replacements.