For example:

Foo id:1
Bar p:1
Baz p:1

Would only store the ID for the last added edge (Foo,Baz) and forget
about the ID of 1 for (Foo,Bar).

Also make sure that add_edge does not add an edge twice, otherwise the
edge ID could be accidentally overwritten.

The function takes care that all edges are present in the dependency
graph.

This fixes a bug that when adding a todo with a parent which appears
later in the file, the edge wouldn't be drawn.

In fact, the id: and p: tags in a todo are edges, so this value has to
be stored somewhere.

Also, adapt run.sh such that every test is automatically run when no
specific test was given as a parameter.

No tests though, not sure how to properly test this. Monkey patching the
datetime module doesn't appear to be working on Linux.

The number represents the position in the backing store, and can be
printed on the stdout to help the user to identify the correct task.

A view is a subset of the todo list where sorters and filters may be
applied.

At this point there are two filters:

* A grep filter that finds a pattern in a todo text
* A relevance filter, which selects todos that are relevant at this
  moment.