New Python Django application default to PBKDF2 with SHA256 as a
password mechanism. This adds support for that mechanism in our
password hasher class. This will be needed in the tests for the new
PDO auth plugin.

for possible upcoming refactoring

This allows to quickly check if a variable is empty but keeps strings
and numbers evaluating to zero.

These tests were passing on travis-ci but failing when run locally because the tests were assuming the value of DOKU_BASE would be '/./' but it was actually '/tmp/' instead.

This makes the entire testsuite run through on HHVM 3.8.0 on phpunit
beta-2015-06-18 :-)

Note: this does not mean that DokuWiki is fully compatible with HHVM,
but it's a good step!

- put test comments in more appropriate spot
- move appending replacement line alongside its search/delete code

The changes should also:
- fix unlikely edge case when replacement line is the same as the
  old line (would have resulted in timeout)
- reduce memory footprint
- avoid applying string search beyond maxlines replacement limit

This was previously carried out in three separate places. Refactor
that code to use the new function.

Update tpl_pageTitle test to use a manager level admin plugin.

add unit test coverage

- replace PHP4 style class constructor function names (based on
  class name) with php 5 __construct()
Also remove some '&' reference operators used with objects
And add some object type hints

io_deleteFromFile is a special case of io_replaceInFile. Since
the io_replace tests check compressed files, it doesn't have to
be done in the io_delete test.

this is far from comprehensible, but should give an idea how the new
library works and how to write tests

this group is skipped by default

it seems that different zlib versions behave different with corrupted files.
Some return false, some return whatever they still can read from the
file. the file now should no longer be readable by any version.

This also reverses the order of crypto protocols tried again. Using TLS
first again. related to #915

The code reading .bz2 compressed files did not correctly check for
possible read errors. In case of a corrupted file this could have led to
an infinite loop.

Thanks to Filippo Cavallarin from www.segment.technology for dicovering
this bug.