[question] Life after split() (P3682)

I can see how to make N-ary tree shaped task graphs using fork_join and when_all.

In the Bulirsch-Stoer integrator, when filling in Neville's table, the graph looks like

graph BT;
00-->10;
01-->10;
01-->11;
02-->11;
02-->12;
03-->12;
10-->20;
11-->20;
11-->21;
12-->21;
20-->30;
21-->30;

I got something working with a pattern of the form

auto job10 = stdexec::when_all(job00, job01) | stdexec::then(fn(10)) | stdexec::split();
auto job11 = stdexec::when_all(job01, job02) | stdexec::then(fn(11)) | stdexec::split();
auto job20 = stdexec::when_all(job10, job11) | stdexec::then(fn(20)) | stdexec::split();

With split() being removed from the standard, what would an equivalent be?

I suppose the underlying question is really: 'How do you create a general task graph?`

The issue centers on the Bulirsch-Stoer integrator's Neville-table graph and the shown stdexec::when_all, stdexec::then, and split() composition; start by tracing those operators and the related fork_join/when_all pattern in the repository. Determine and document the supported post-split() way to express the general task graph.