Enable caching of pipeline steps

[tarekoraby]

Quite often, pipeline steps don't change between runs. For example, a pre-processing step on the same input dataset will always generate the same processed output. In those cases, it would be beneficial not to waste time and resources re-executing the step. And while the Valohai UI offers an option that allows users to manually reuse steps from previous runs, this manual option

1. can be missed by less experienced users
2. doesn't help in cases where the pipeline should be automatically triggered.

Thus, it would be nice to have the option to enable caching on specific steps. When this caching is enabled, Valohai should try to find a previous execution of the same step that has the same inputs + parameters. If such execution is found, the output from this previous execution will be reused rather than re-running the step again.

However, there might be situations where the step should be cached most of the time, except that it should be occasionally rerunning the whole step. For example, a pre-processing step might be programmatically retrieving data from a remote API. Most of the time, this remote data remains unchanged. But occasionally (say once a week), the API should be rechecked for updates. In such a scenario, it would be essential to offer the option to force run a step, even if caching is enabled for that step.