Conflicts:
	setup.py

Also marked a few functions private.

Move format parsing logic to ListFormatParser

This way we can safely remove `PrettyPrinterIndentFilter` and output
length is still properly maintained by `ListFormatParser`.

ListFormatParser is now the core of PrettyPrinterFormatFilter, but can
also be easily used in another areas. Whole functionality of
PrettyPrinterAlignFilter was also moved into it.

Main advantage of having separate parser for list_format is possibility
of initializing it and resolving placeholders only once. Previous
implementation did that with each Todo object.

- From now on literal `%` signs always have to be escaped with `\` (`\%`),
even inside conditional braces.
- `%%` won't work anymore.
- Nonexistent placeholders (for example`%&`) will be expanded to empty
  strings (`''`).

This change fixes also some bugs:
- inappropriate output width when \t was used without %S
  (example: `%p{ }\t%{ }p`)
- curly braces inside conditional braces (example: `%{{}p{}}`)

Minor: recognize literal `\t` in list_format config option and in `-F`.

They mostly deal with corner cases (small terminals, dodgy texts in the
prefix/suffix).

Also prepare the test for the new syntax where a literal % will be
written as \% instead of %%.

This corresponds to the change made in commit 33794435.

Update tests for ls_format

Test ignore_weekends clause in Importance

2015-11-06 as freeze_time date was introduced in #75 so let's stick with
it also here.

Also introduce freezegun.freeze_time decorator.

Update tests for Importance

2 more tests of calculation importance from due date.

This increased (branch) coverage to 100%. A few corner cases were
missing with strange default commands in the configuration.

This was missing according to coverage.

It has been replaced by the %k keyword when using formatting (`ls -F`).

Conflicts:
	setup.py

Add tests for list_format

Also add freezegun to dependencies -  we use it for mocking datetime
objects.

Users should rather use %k in list_format if they want to avoid hidden
tags on the output.

Coveralls doesn't support branch coverage, while codecov does.

A couple more tests

Thanks to @mruwek for reporting.

The project/context, tag and URL matching patterns are not aware of any color codes that are already prepended to a word. This affects items that appear at the start of a line.

Given a todo string where a priority color was already added:

\x1b[0mkey:value

Then the match would turn it into something like this:

\x1b[\x1b[1;32m0mkey:value\x1b[0m

So you get an ugly 0m prefix to the key:value text, because ] is a word boundary and 0m are non-whitespace characters that end up as being part of the key.

Report and suggested fix by @mruwek. It became apparent with `ls -F %k`, which puts tags at the start of a line, a not so common case that slipped through when writing the colorization code.