wasm optimization pass in mops (build + bench)?

[Kamirus]

Optimization (ic-wasm optimize) isn't done by mops today — dfx did it silently via optimize: "cycles". On pocket-ic nothing optimizes, so bench numbers ≠ deployed numbers (sha256: ~163M raw vs ~134M optimized). Open question before any design.

Aspects to settle:

• Scope: does mops produce the optimized deployable artifact, or just optimize for bench fidelity?
• Consistency: build, bench, and deploy must optimize the same way; never diverge silently.
• Ownership vs icp-cli recipe: recipe already does shrink/compress; avoid stacking/conflicting passes.
• Config: flag vs mops.toml: config only justified if shared across commands.
• ic-wasm versioning: pin it as a toolchain tool, matched to icp-cli?
• Default: opt-in vs opt-out (changes everyone's numbers).
• Visibility: applied level/tool/version must show in output.
• Blocker: ic-wasm optimize fails on moc ≥1.10 (multivalue, no target_features section): needs moc fix or mops injecting the section.