fix(runtime): explicitly load previous results

[eddiebergman]

• closes Error if I rerun an experiment that uses hyperband as a searcher #103
• closes Error caused by post_run_summary argument if using hyperband as a searcher #104

Test by rerunning the reproducible examples (Many thanks @danrgll!)

---

The issue was from my previous re-working runtime, where I forgot to ensure that if there was a trial (i.e. config_1_1) that relied on a previous trial (config_1_0), to ensure that it was loaded into that cache before config_1_1 was loaded in. Main fix was just to sort accordingly and then process each one 1-by-1.

        # NOTE: We sort all trials such that we process previous trials first, i.e.
        # if trial_3 has trial_2 as previous, we process trial_2 first, which
        # requires trial_1 to have been processed first.
        def _depth(trial: Trial.Disk) -> int:
            depth = 0
            previous = trial.previous_config_id
            while previous is not None:
                depth += 1
                previous_trial = _disk_lookup.get(previous)
                if previous_trial is None:
                    raise RuntimeError(
                        "Previous trial not found on disk when processing a trial."
                        " This should not happen as if a tria has a previous trial,"
                        " then it should be present and evaluated on disk.",
                    )
                previous = previous_trial.previous_config_id

            return depth

        # This allows is to traverse linearly and used cached values of previous
        # trial data loading, as done below.
        _disks.sort(key=_depth)

        for disk in _disks:
            # process the trial

---

PR also attempts to simplify things a bit and reduce some duplication, as fixing this issue would have introduced a lot of spaghetti.