Support recovery from failed task dependencies

[lefou]

Currently, tasks can only depend on the success value of other tasks. To depend on another task result, we use the def apply(): T method.

There are situations, where task failure does not necessarily means a build failure.

Here are some made up examples:

• A framework-independent test reporter, which should count succeeded/failed test stats, even when the test target failed
• an build failure analyzer plugin may want to analyze compile errors, by depending on compile tasks
• the prepareOffline task may want to fetch as much dependencies as possible, even when some are going to be unsuccessful
• we need to work around some issue in another plugin without rewriting the other plugin
• there are probably more and better scenarios...

Thoughts about the implementation:

• We need a new accessor to depend on the raw task result Result[T], not just the sucess value Result.Success[T](value), e.g. def raw(): Result[T]
• When creating the evaluation plan, we need to also track all raw dependencies (raw inputs)
• When a task fails, we need to check whether there are any unprocessed raw dependants (raw consumers)
• If there are unprocessed raw consumers, we kind-of need to keep going until we have processed all raw consumers
• If a task has inputs to other failed tasks we fail it as before, but if all failed inputs are consumed via raw inputs, we run the task
• If a task fails, that has no raw comsumers, we can fail the build (when fail-fast) or continue (when keep-going).

[lefou]

For some tasks like test task, we might want to also provide a (different) failure result (type), so that it is possible to better process (and recover) a failure.