perf: cache scheduler depth prefixes (#868)

* bench: cover deep scheduler waves

* optimize(scheduler-depth-prefixes): memoize every ancestry prefix visited while computing wave depths, then unwind each unknown path iteratively

* optimize(scheduler-depth-prefixes): build a batch-local queued-block set and resolve depths only until the nearest queued ancestor

* test: guard scheduler depth ordering and scaling

* bench: resolve vite from a workspace fixture

---------

Co-authored-by: Dominic Gannaway <trueadm@users.noreply.github.com>