Adds our own CoM callback

[jackbdoughty]

We could not use Bluesky's centre of mass callback as it was not suited for the way we would want to use it, so this is our implementation for one.

To test:

• Make sure that all tests pass.
• Run the following scans and check that you get sensible CoMs for each. (Send me a message if you need to know how to set up to scan blocks against eachother / create fake data):
  • A scan with non constant point spacing.
  • A scan where some of the values are negative / have a background.
  • The "there-and-back" scan so x points would be like: -2 -1 0 1 2 1 0 -1 -2. (Non continuous points)

[rerpha]

we should use this in our standard callbacks class - if this ticket is merged before #104 i will add it in, if not i think this PR should

[jackbdoughty]

#108 Will fix the above errors

[Tom-Willemsen]

I'm going to push some stuff to a trapeze branch and we should probably chat tomorrow... I found some extra-fun edge cases that I hadn't previously considered...

e.g. this case:

def test_tricky_non_constant_spacing():
    ps.start({})

    ps.event({"data": {"x": 0, "y": 1}})
    ps.event({"data": {"x": 0.1, "y": 1}})
    ps.event({"data": {"x": 4, "y": 1}})
    ps.event({"data": {"x": 5, "y": 0}})
    ps.event({"data": {"x": 6, "y": 1}})
    ps.event({"data": {"x": 10, "y": 1}})

    ps.stop({})

    assert ps["com"] == 5.0

[Tom-Willemsen]

Sorry I know I've gone back and forth on this, but I'm not fully convinced it's a good idea to subclass _DefaultPeakStats here actually - the upstream one has __slots__ defined and some other potential traps.

Maybe here it is better to just make a standalone callback... subclassing CollectThenCompute or similar.