Updating scheduling algorithm to "stable best guess sequencing"

[justinbmeyer]

The autoscheduler currently plans out the "long pole" first and then fills out the gaps with other epics.

This makes sense if you know how long everything is going to take and what you might be able to "squeeze" in.

However, folks making plans would probably need to make a decision and "stick with it".

For example, say there was an A.1 epic that blocked A.2 and A.3.

== EPIC A.1 == == EPIC A.3 ==
= EPIC B =     == EPIC A.2 ==

If I thought Epic B would "fit it", I'd start it right away.

What happens with the probablistic estimator ... is that it sometimes knows Epic B won't fit it, so it won't schedule it.

I think the scheduler should be deterministic here too. If we were going to schedule "B", we should always schedule it there. If it moves "Epic A.2" out, so be it.

This would probably match folks expected scheduling more accurately and make the tracks results make more sense.

Implementation

I think the scheduler will need to do a run based on the "confidence inflated" interval ...

Then it should sequence epics by that sequence's "start date". We should still "fill in" any gaps, push as far "left" as we could.