The B3 hook's unlock cascade was compare-mode only, so a replace run left the
completed node unstamped and no tech unlocked -- and its only per-call counter was
gated on the cascade, so a replace run had no instrument of its own saying a
completion had happened at all. Lane CR needs both halves to ask whether
ProcessResearch can be displaced.
research.replace_cascade=on (default off, so the shipped behaviour is unchanged and
still measurable) lets a replace-mode call write the four TechNode words
SetResearched stamps and the tree's completion-order counter. The ServerPlayer half
of the callback stays unmodelled and stays declared: no event is posted, no
ObservedTech element is appended, no tech effect is applied. The pair of settings is
the measurement -- same binary, one config line, and the leaf difference between the
two autosaves is the cascade's own contribution.
The per-call log line now prints in every mode and counts completions from the pass's
own step results, so 'a completion fired' is a statement by the instrument rather
than an inference from the artefact under test.
