v0.1.0.dev9

What's Changed

• fix(optimizer): metric-correct Riemannian Adam second moment by @smaniches in #84
• fix(sheaf): guard explicit-Euler diffusion against unstable steps by @smaniches in #85
• feat(examples): reproducible end-to-end SO(3) aggregation demonstration by @smaniches in #86
• ci: update scipy requirement from <1.16,>=1.10 to >=1.10,<1.18 in the runtime group by @dependabot[bot] in #87
• test: bridge actual point-action aggregation to equivalent neural networks by @smaniches in #88
• fix: preserve committed federated-round state on failure by @smaniches in #89
• research: publish byte-exact six-seed neural gauge evidence and replay verifier by @smaniches in #90
• docs: reconcile real-neural experiment with core API validation boundary by @smaniches in #91
• ci: stop masking benchmark-test failures by @smaniches in #92
• release: publish metric-aware numerical corrections as 0.1.0.dev9 by @smaniches in #94

Full Changelog: v0.1.0.dev8...v0.1.0.dev9