Refresh multi-organism validation on the deterministic placement (stacks on #83)

[edkerk]

Completes the multi-organism validation for the deterministic placement tie-break in #83.

Stacks on #83 (feat/placement-tiebreak) — merge #83 first; this PR's diff reduces to just the multi-organism doc update once #83 lands.

#83 fixed the reaction-placement co-optimum degeneracy but only re-measured yeast, leaving the Human-GEM / AraCore / iCre1355 reaction-agreement figures in multiorganism_validation.md from the old arbitrary placement (footnoted "await re-measurement"). This re-runs benchmark_certified_multiorg.py on the tie-break code and refreshes those rows.

Result

organism	reaction agr	gene agr	transports	blocked
yeast-GEM	72.0 % → 72.5 %	88.7 % (=)	1001 → 967	29.7 % (=)
Human-GEM	52.8 % → 56.9 %	79.4 % (=)	3701 → 3608	10.7 % (=)
AraCore	81.8 % → 82.1 %	93.7 % (=)	215 → 200	0.0 % (=)
iCre1355	53.4 % → 54.1 %	70.5 % (=)	1108 → 1083	19.0 % → 32.9 %

• Reaction agreement rose or held everywhere; Human-GEM's +4.1 pt is the largest move (AraCore and iCre1355 were already near their ceilings, so the tie-break mostly confirms them).
• Gene agreement is unchanged per model — direct confirmation, across four kingdoms at genome scale, that the tie-break fixes only the reaction placement and leaves the gene layout untouched.
• Fewer transports across the board — coherent placement fragments fewer metabolites.
• One non-obvious shift: iCre1355's blocked fraction rose 19.0 % → 32.9 %. Tighter, less transport-heavy placement leaves more reactions dead-ended (the model still certifies). Reported honestly — the original task assumed blocked was placement-invariant; it isn't, because placement determines connectivity.

Also removes the "await re-measurement" footnote and bumps the prose Human-GEM reaction count to the on-disk model version (12854 → 12877). All three organisms certified; measured with cobra.Configuration().processes = 1 for Windows find_blocked_reactions safety; reproducible run-to-run.