pure fusion in Dynamic

[zyla]

This PR modifies Dynamic by adding a special-case constructor for constants (DynPure). Constants are propagated, for example map on a constant also produces a constant.

This enables some optimizations:

• reduces the number of nodes in the Incremental graph
• reduces the number of subscriptions and unsubscriptions (subscribing to a DynPure is a no-op)

This can help especially for computations which are expressed as Dynamic but where the actual dynamism is limited.

[erykciepiela]

It's interesting: Dynamic Apply instance doesn't need this, Dynamic Bind instance does. And mapAsync does. It seems this should be avoidable so we have even less nodes. Or it isn't?

[zyla]

Interesting observation!

In principle we could specialize bind and mapAsync so that we don't need it. It's just that in these cases I think we don't gain much, and it would require substantial work.

Firsy, in both cases the Pure has no chance of propagating further.
mapAsync on a Pure still has to return a DynNode, to handle the loading state.
And for RHS of a bind case: if the LHS is still dynamic, we can't propagate anyway. And we already handle the case where the LHS is Pure.
So this would be a small local gain. Async is usually so heavy that the specialization wouldn't make sense. For bind I think we can still consider it.

Second, in the current shape rhe change doesn't need to touch the Incremental layer, which buys us a but of "obviously correct"ness.

[kozak]

LG!