feat(bench): EOP by default + --query for query-context measurement

[alexandru-uta]

mops bench compiled benchmark canisters with --legacy-persistence and ran every cell as an update call — neither reflects how a real query canister executes on the IC. This makes the instruction counts unrepresentative for read-only/query workloads (they include GC work a query never does, under the wrong persistence mode).

Changes

• EOP by default — getMocArgs no longer forces --legacy-persistence (it was gated only on moc ≥ 0.15). The bench canister is already a persistent actor, so dropping the flag compiles it under enhanced orthogonal persistence — moc's default and the mode real canisters run.
• --query — measures each cell via the canister's existing runCellQuery (a query method) instead of runCellUpdateAwait. Queries run no GC on the IC, so the counts exclude GC work an update would incur. Synchronous runners only (async/inter-canister runners still need the update path).
• Verbose pipeline now prints context (query/update) and persistence.

No change to the mops-bench library — the canister template already exposes runCellQuery.

Validation

• npm run check (tsc) passes.
• Smoke-tested via tsx against a real project: verbose prints context: query / persistence: enhanced / gc: incremental, and the measured numbers shift as expected — the real-runWith floor drops sharply (no GC counted) while raw allocation rises slightly under 64-bit EOP.

Docs / changelog

• docs/docs/cli/4-dev/02-mops-bench.md: updated "How it works" + new ### --query section.
• cli/CHANGELOG.md: entries under ## Next.
• --help carries a description for --query.
• The mops-cli skill documents no individual bench flags, so nothing there to update.

🤖 Generated with Claude Code

[github-actions]

Cursor AI review

👍 APPROVE — looks safe to merge

Category	Assessment	Details
Summary	✅	Switches mops bench to EOP-by-default (was forced --legacy-persistence on moc ≥ 0.15) and adds --query / --legacy-persistence flags so measurements can match real query-canister execution; docs, changelog, and --help updated.
Code Quality	✅	Minimal, focused diff reusing existing runCellQuery in the bench canister template; no speculative helpers.
Consistency	✅	Commander options wire cleanly into BenchOptions; sibling docs/changelog/--help stay aligned per project rules.
Security	✅	Local dev/benchmark tooling only; no auth, registry, or signing surface touched.
Tests	✅	No bench Jest coverage pre-existed; change is localized and author smoke-tested; no redundant tests added.
Maintainability	✅	Clear comments and verbose pipeline output for the new modes; aligns with existing TODO to drop legacy-persistence default.

Verdict

Decision: APPROVE
Risk: Low
Reason: The behavioral changes are intentional, well-documented, and provide --legacy-persistence as an opt-out; implementation correctly gates the persistence flag on moc version and routes --query through the existing canister query method without touching production CLI/registry paths.

---

Generated for commit 6c3d9aa