The closest analogue I can find to Bergson's idea of "duration" is a topological neighborhood as opposed to a geometric point. In topology, sometimes one canot provide an exact location of a point in terms of coordinates but instead can only specify a set of neighborhoods it belongs to (and sometimes an infinite series of neighborhoods zeroes in on a distinct location.)
Bergson's durations are such neighborhoods: within them, time does not cleanly partition into identifiable moments but instead is a bunch of stuff happening in nebulous flux, and this allows for some concept of contingency rather than simple deterministic causality.
How does this square with what we sscientifically know about time? Einstenian relativity is chronological, it accounts for the synchronization of clocks and the motions of celestial bodies, and in doing so tells us something undeniably true; but it seems very presumptuous to say this is the only relevant way to conceive of the idea of time.
Many processes can only be thought of in terms of trajectories and not specific moments: stories, seasons, lifecycles. No, they do not ultimately "reduce" to chronological time, it's just different math. This is not about physics vs "metaphysics", it's about not using a hammer on a screw just because the hammer works for a lot of other things.
Technical addendum: one can imagine chronological time as being comprised of points that are converged on by infinite series of durations (basically the real number line.)
But seen this way, you could imagine plenty of "stuff" "outside" of this topology. Perhaps you could even call these less Euclidean topologies "spirit" in some sense.
By the framework of time one sees in Deleuze's Difference and Repetition one could even imagine habit and memory as two different topologies, the former treating the present as something open and the past closed, and memory vice-versa, though I am really shooting from the hip here.