Faster BeliefDataFrame with cached properties

[nhoening]

There are some properties on BeliefsDataFrame which seem to be re-computed every time they are accessed (example below).

Would it be a good idea to cache them?

    @property
    def knowledge_times(self) -> pd.DatetimeIndex:
        return pd.DatetimeIndex(
            self.event_starts.to_series(name="knowledge_time").apply(
                lambda event_start: self.sensor.knowledge_time(
                    event_start, self.event_resolution
                )
            )
        )

Maybe there are some inspirations for how this should work to be found in pandas (maybe grep @property and/or lru_cache).

And we need to check if we can guarantee that we are creating new BDF objects with every operation that expands or reduces the number of rows. It looks likely, even append_from_time_series returns a new BDF object. Or alternatively, invalidate all caches in other cases.

[nhoening]

@Flix6x Is this still a candidate in our search to generate speed improvements?