[Q] How can a trial access results of previous trials from the same experiment?

I know that one can use the `ExperimentAnalysis` (from `ray[tune]`) to load the data from a previous session, but I'm curious about how one would be able to do that from within a `ray` process (it's called a `raylet`, right?) running a trial, especially since there are some concurrency considerations that need to be taken into account, e.g., the trial might decide to read the history while a different process is writing to that experiment directory (in its own trial subdir).

I don't think it's particularly relevant for the question, but I'm asking this in the context in which I'd like to try using the `flaml` scheduler and, if the scheduler decides to keep running a trial after it has already run for a while, it would be nice if it would be able take into account previously reported performance scores instead of starting from scratch and report the exact same initial progress as before. Note that this only works is a model-less and deterministic setting, though.