Support fallible tasks with graceful continuation

[cassaundra]

When a fallible task fails, note the result and continue execution through non-dependent tasks. Possibly add an on-failure hook (or similar) as well.

This is a precursor to making quake into a task runner (see #30).

Note: in watch mode (#14), the intended behavior for task failure is to abort all dependent tasks of those that failed and re-run them when a file change is detected. This is a separate behavior (at least conceptually) from this feature.