gc-matrix: the 'moved' liveness counter sums the C4b mark-sweep evacuation with the copying minor, so a move-arm can be green with zero scavenged objects

[proggeramlug]

Found while re-measuring the matrix after #7019. Small, but it silently weakens the one thing this gate exists to assert.

The problem

scripts/gc_repsel_matrix.sh derives its moved= evidence — the liveness predicate for every requires=move arm — by summing two different collectors' counters into one number:

moved=$( { grep -oE 'moved_objects=[0-9]+' ... ; \
           grep -oE '\[gc-copy-minor\] ran copied_objects=[0-9]+' ... ; } \
         | grep -oE '[0-9]+$' | awk '{s+=$1} END {print s+0}')

• moved_objects= comes from the C4b evacuation policy inside the mark-sweep collector — the pre-existing non-moving-minor path that relocates tenured objects during a full GC.
• [gc-copy-minor] ran copied_objects= comes from the evacuating young-gen scavenge, which is what gc: default-on evacuating young-gen scavenge (moving minor at precise safepoints) #7019 made default-on and what the move arms are meant to certify.

A cell can therefore report moved=7 610 512 and be marked live for a move arm while the copying minor never ran once.

It is not hypothetical — it is the current state of the default arm

--arms all --pressure 8 on origin/main @ 4340bffc9, 22 rows, counters separated by hand:

arm	Σ copied_objects (scavenge)	Σ moved_objects (C4b)	harness's moved
default	0	7 610 512	7 610 512
verify_evac	0	7 301 552	7 301 552
evac_minor	3 628 777	0	3 628 777

The first two rows read as "movement happened" in the arm-liveness summary. No object was moved by the collector under test. (The reason those arms run no scavenge at all is filed separately — it is the pressure knob colliding with the moving-defer hard cap.)

Suggested direction

Report the two counters as separate columns and let each arm state which one satisfies it. A requires=move arm added for the evacuating scavenge should require copied_objects > 0 specifically; moved_objects > 0 is a fine liveness signal for a full-GC evacuation arm, but it is a different claim.

While there: the collect predicate has the same shape of weakness, already filed as #7017 (a teardown-boundary cycle that frees nothing is indistinguishable from a mid-program one).

Related: #7019, #6942, #6946, #6950, #7017.