No dependency tracking is needed and gives a huge performance
improvement for huge done.txt files.

The subclass implements the dependency tracking of todos, which can be
quite intensive for huge todo.txt files, and is not always needed. For
instance, dependency tracking is not needed for archiving.

Since they're looked up quite often, they should be cached insteadof going through the ConfigParser machinery.

Most of the time was spent in the todo_by_hash function which is
not efficient enough for large todo files.

This optimizes for large todo files (or large archive files). For each
added todo item, _update_parent_cache was called N times, which in turn
calls todo_by_hash which also does N iterations, where N is the number
of todos added at that point. This doesn't scale with a couple of
hundreds of todo items.

Instead, introduce a list variant of add_todo which does the cache
update only once.

It's possible to use <, <=, =, > and >= to select todo items based
on a integer or date tag.

When using negation, the "--" separator should be used, so getopt
doesn't stumble over -word constructs.

Also add a test case that checks functionality of a negation with
multiple words.

There are actually two force modes, one to force adding a tag and
one force mode that omits user interaction. The first kind is done
with -a, the second one with -f.

The meaning of -f changes in this commit, but now it's consistent
with the other subcommands.

The default value for this limit is -1 now. Any non-negative value
will be used for the LimitFilter.

Strict recurrence means that the actual due date of the todo item is
used, instead of using today's date.