Refine Diamond benchmark estimation and input-injection accounting

[SoraSuegami]

Summary

• Split shared input-injection benchmark logic into a dedicated estimator and reuse it from Diamond iO and Diamond WE
• Remove compatibility-only constructors and make the test scaffolding construct explicit unit-cost inputs
• Add missing artifact decode, materialization, and serialization costs so benchmark estimates better match the CPU and GPU execution paths
• Clean up related simulation code and GPU test coverage to keep the estimator and runtime paths aligned

Testing

• cargo +nightly fmt --all
• cargo check
• cargo test --lib diamond_we_bench_estimate
• cargo check --features gpu

[chatgpt-codex-connector]

💡 Codex Review

mxx/references/diamond_io.pdf

Line 1 in 830d5cb

%PDF-1.5

Keep generated artifacts out of references

AGENTS.md states that directories named references are read-only for humans/agents to consult and that agents must never edit them. This commit adds references/diamond_io.pdf, so an agent-produced change with this file violates the repository workflow; please remove it from the patch or place the artifact somewhere outside references.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".