doc: reconcile the peek execution design with what the stack now does

The document sat on top of the stack so it could be read against the tree
below it, and a review against that tree found claims it does not support.

Corrected: a quotation attributed to the code that exists nowhere; a fifth
cancellation state that is not a state, since a cancellation landing mid
inline slice arrives in one of the other four; a retention bound stated
for every scan when it holds only for one that may use the stash; content
of the merged fueled-iterator layer that was never in it; "fewer
`PendingPeek` states than today" where the count is the same; a deleted
spare cursor that belongs to an unmerged proposal rather than to this
stack; a deleted user-facing metric the Deleted list omitted; a cursor
setup timer described as summing slices and mixing substrates when it is
one measurement taken on the worker; and present-tense descriptions of
code the layers below have already replaced.

Updated for the stack as it stands. Every parameter is replica-scoped now,
so the argument for pinning three of them environment-wide is gone: the
two paths are required to agree, and where they do not that is a defect to
fix rather than a reason to deny per-replica targeting. The permit bound
is a nano-fraction of a runtime's workers rather than a count with a
sentinel. A peek waiting on its turn keeps the worker from parking rather
than asking timely to wake it. Fuel is charged only for positions a walk
inspects, and the charge is independent of how the walk is sliced. The
result-size ceiling now bounds a stashed answer across the workers that
produced it, not just each worker's share. The stash transition ships
default-on and is outside the kill switch's rollback, which the Kill
switch section now says.

The measurements name where they came from and that they are one-off
staging observations, and the diagram shows the exits the walk actually
has rather than two of them.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: antiguru