perf(bench): make snap-then-simulate benchmarks faster

[gilcu3]

Closes #2774
takes 12 minutes on my machine:

❯ MAX_MALICIOUS=40 cargo bench -p threshold-signatures --bench advanced_dkg
    Finished `bench` profile [optimized] target(s) in 0.16s
     Running benches/advanced_dkg.rs (target/release/deps/advanced_dkg-433c4cf2c869f36c)
dkg/dkg_secp256k1_MAX_MALICIOUS_40_PARTICIPANTS_41
                        time:   [2.5496 ms 2.5653 ms 2.5822 ms]
Found 3 outliers among 15 measurements (20.00%)
  3 (20.00%) low mild

dkg/dkg_ed25519_MAX_MALICIOUS_40_PARTICIPANTS_41
                        time:   [2.8598 ms 2.9045 ms 2.9366 ms]
Found 2 outliers among 15 measurements (13.33%)
  2 (13.33%) low mild

dkg/dkg_bls12381_MAX_MALICIOUS_40_PARTICIPANTS_41
                        time:   [2.8873 ms 2.9659 ms 3.0107 ms]

[claude]

PR title type suggestion: This PR changes only benchmark and test utility files, so the type prefix should probably be test: instead of perf:.

Suggested title: test: make snap-then-simulate benchmarks faster

[claude]

Code Review

No critical issues found. The refactoring correctly hoists expensive one-time protocol setup out of Criterion per-sample iter_batched closure, caching the Simulator and cloning it cheaply for each iteration. Key observations:

• The ProtocolSnapshot signature change (self to &self on new and get_received_messages) is safe since get_received_messages already delegates to read_all_messages which only borrows.
• MockCryptoRng clone preserves determinism identically to the prior approach (same seed produces same state each run).
• derive Clone on Simulator is appropriate since it only contains a Participant (Copy) and Vec.

Approved

[claude]

PR title type suggestion: This PR changes only benchmark and test utility files, so the type prefix should probably be test: instead of perf:.

Suggested title: test: make snap-then-simulate benchmarks faster