Expression-mode for del, depri, do, pri, postpone.

- add test for direct todo.txt editing
- remove unneceseary `u()`

- execute_multi_specific() -> _execute_multi_specific()
- execute_not_multi() -> _execute_not_multi()

MultiCommand._catch_todo_errors() prevents occuring of those.

This required some refactoring in MultiCommand and ExpressionCommand.

Whole part about processing flags was refactored. Code related to flags
from DCommand was moved into MultiCommand and reused from there in all
"multi" commands.

New attribute: ExpressionCommand.last_argument. Should be set to True if
command is using the last argument as something other than todo ID or
part of an expression (as in case of postpone or pri). By default set to
False.

New method: MultiCommand.execute_not_multi(): wrapper for functionality
not connected with todo IDs (like `topydo edit` or `topydo edit -d`).

New attribute: MultiCommand.multi_mode: Should be set to True if user
wants to operate on multiple todo items. True by default.

New method: MultiCommand.get_todos_from_expr(): gets todo IDs based on
expression and puts it in MultiCommand.todos.

EditCommand._open_in_editor() now accepts only one argument - file to
editing. It handles now also OSError exceptions. It is used also to
opening todo.txt and done.txt.

MultiCommand.get_todos() doesn't accept anymore todo IDs as an argument.
Instead it's using inherited Expression.last_argument` to decide upon
where to find todo IDs.

Now desired result of DCommand.get_flags() should be more apparent.

Now every call to editor is done by `EditCommand._open_in_editor()` and
whole process of getting editor name is also done there.

When a subcommand finishes, _post_excecute() is called for sorting,
archiving and writing the file. But sorting is also a subcommand, which
in turn also calls _post_execute(), etc, etc.

So let the caller of _execute() decide when to call _post_execute.

This fixes issue #46.

Just make sure that proper arguments are passed to GrepFilter in the
first place.

This started as a refactoring of the View class: to get rid of the
pretty_print() function and anything releated to printers. So the View
class now solely focusses on sorting and viewing a bunch of todos. It's
up to the users of the View class to retrieve the todos in the view and
pass them on to a Printer instance.

The View class was quite involved with the proper Unicode support that
was added lately. There were still encoded string passed through this
class rather than Unicode strings. Getting rid of this affected
basically all other places in the codebase.

The result is that there are less string conversions; printed todo items
stay Unicode as long as possible, until they're actually written out to
standard output or to a file. I got rid of one text_type() call from the
six library, all @python_2_unicode_compatible decorators are gone and
the tests no longer need the utf8() function because everything is
Unicode now.

Remove utf8 function.

No need to toggle the highlighting of projects and contexts anymore. If
you want to turn it off, simply assign an empty value to the
project/context colors:

  [colorscheme]
  project_color =
  context_color =

Thanks to Jacek for pointing out.

Safe 16 base colors

- Colors specified by name will return safe ANSI codes.
- Colors specified by number will return ANSI code from xterm 256 color
  chart.
- Defaults are also *safe*.

This is a revert of commit fbed527f, which resulted in too many items
after a project/context being highlighted.

Instead, don't print a NEUTRAL_COLOR code at the end of the todo, but
only when we finish applying all the color codes. This makes sure that
the regexp to match a project/context also works at the end of a line.

Also removed some logic to remove multiple sequences of NEUTRAL_COLOR at
the end of the line, this is not necessary.

@\033[0m gets matched by '\B(\+|@)(\S*\w)', and as a result, the @ would
be colored to the context color.

Probably an indenting mistake.

Add possibility to configure colorscheme of topydo

The completer would be instantiated before the todo list was loaded from
the file. Now, the completer is instantiated every time the todo list is
loaded.

+ -> @

- Add new module `topydo.lib.Colors` and class `Colors` for handling
  output colors and their ANSI codes.
- Create new config options with colors for priorities, projects,
  contexts, metadata, and links and place them under "colorscheme"
  section. When certain config option is absent default color code will be
  applied.
- Empty values:
    ! priority_colors will fall back to NEUTRAL_COLOR
    ! project_color, context_color, link_color, metadata_color will
      return empty string, so priority_color (if configured and applicable) of
      specific item should be used
- Return NEUTRAL_COLOR if supplied color number is out range
- Support user-friendly color names in config. At least for 16 base
  terminal colors.

Instead of creating a subcommand for every possible output format we're
going to think of in the future, just let ListCommand handle the output
format by connecting to the right format printer.

In this commit, the 'ical' subcommand is deprecated, to be removed after
the next release.

The 'json' subcommand has been removed altogether, it existed for only a
few hours.

The todo.txt file can be modified while the prompt is running. Make sure
to reload the file when the mtime has changed since the previous load.

This addresses the issue that changes after the 'edit' subcommand were
not visible when returning to the topydo prompt (issue #33).

Also, other changes done while the prompt is running will be picked up.

A safety measure was taken to refuse executing a command when the
todo.txt file was modified while the prompt was visible (i.e. the user
was busy typing a command). By doing so, subcommands such as 'do' and
'del' will only operate on the actual state of the todo.txt file, not
some previous state still in memory.