hover edge cases on period positioned points

[archmoj]

Clear improvement. There are three funny cases I want to point out, though we may choose not to do anything about them:

(1) the only place I can put the mouse and NOT get a label for any bar is slightly to the left of a first-of-month scatter point:

If I put the mouse to the right of this point I do get a bar hover:

This is presumably because the mouse is not over the range of the bar that matches the winning scatter point - so we include the previous bar in our original set of hovered points but then disqualify it because it's inconsistent with the scatter point.
This happens at all month boundaries in x mode, but if I switch to x unified mode, I see this behavior on the Dec/Jan boundary:

but not on the Jan/Feb boundary, there I always get the Feb bar no matter where my mouse is:

(2) When I zoom in enough, there are gaps between scatter points so that none of them are hovered, and only a bar is shown in the hover info. But if I set xperiod to 5 days on the scatter trace, I would have expected that no matter how much I zoom in there would be a winning scatter point, because I'm always over the period of one of them. That doesn't happen, in either x or x unified hovermode. In both modes there's still a gap where I only see the bar's hover.

(3) Except in a very narrow set of conditions, when no matter which scatter gap my mouse is in, the middle scatter point is included in hover data with the bar. I only see this with an xperiod on the scatter and in x hovermode - not with instant-positioned scatter points and not with x unified, and at least in the codepen given there's a very narrow zoom range where this happens, immediately after gaps in the scatter have opened up but before they get too big. Here my mouse is on about Jan 27:

(This third point may become moot if we choose to address the second point)

Originally posted by @alexcjohnson in #5618 (comment)

[alexcjohnson]

Closed by #5683