Rework benchmarks

[l0r1s]

Summary

Converts the benchmark and weight system to the standard FRAME WeightInfo pattern. This lays the groundwork for using parameterized weights where extrinsic cost scales with input size (e.g., number of contributors, batch size), which was not possible with hardcoded values.

Before: Extrinsics had Weight::from_parts(X, 0) hardcoded directly in #[pallet::weight(...)] annotations. CI parsed benchmark stdout and patched values in source files.

After: Each pallet has a weights.rs file, a WeightInfo trait, and extrinsics reference T::WeightInfo::fn_name(). CI uses a syn-based Rust tool to compare old vs new weight values with a threshold.

Note: we generate weights using --runtime + --genesis-builder flags (not --chain) because the runtime uses custom host functions that require the full node binary.

Pallet changes

• pallet_subtensor — 51 extrinsics converted to T::WeightInfo::fn(), 18 without benchmarks kept hardcoded. Removed #[benchmark(extra)] from register_leased_network and terminate_lease so they run in normal benchmark passes.
• pallet_admin_utils — 55 extrinsics converted, 19 kept hardcoded
• pallet_drand — all 3 extrinsics converted
• pallet_shield — all 2 extrinsics converted
• pallet_commitments — all 2 extrinsics converted, stale weights.rs fixed (set_rate_limit renamed to set_max_space), benchmark test suite restored
• pallet_crowdloan — weights.rs regenerated (cleanup, no functional change)

For each pallet: added pub mod weights, type WeightInfo in Config, wired SubstrateWeight<Runtime> in runtime, added type WeightInfo = () to all test mocks.

Benchmark fixes

• pallet_subtensor_swap — fixed remove_liquidity (HotKeyAccountNotExists) and modify_position (SubtokenDisabled) benchmark setup. Added benchmarks for add_liquidity, toggle_user_liquidity, and disable_lp.
• pallet_registry — fixed set_identity and clear_identity benchmarks failing with FundsUnavailable (replaced set_balance(max_value) with mint_into, set identity on caller instead of separate account).

New tooling (support/weight-tools/)

Lightweight crate (~4s build, no runtime dependency):

• weight-compare — parses two weights.rs files with syn, compares base weights (threshold-based), reads/writes (exact), and per-component slopes for parameterized weights. Used by CI.

Scripts

• scripts/benchmark_all.sh — rewritten to generate weights.rs via --output/--template, supports single-pallet runs and SKIP_BUILD=1
• scripts/benchmark_action.sh — rewritten to use weight-compare with 40% drift threshold
• Removed 4 redundant per-pallet scripts/benchmark.sh files

CI

• run-benchmarks.yml — cleaned up, extracted repeated skip-label check into .github/scripts/check-skip-label.sh, added weight-compare build step

Other

• .maintain/frame-weight-template.hbs synced with upstream polkadot-sdk
• docs/benchmarks-and-weights.md — guide covering new pallet setup, adding extrinsics, parameterized weights, CI workflow

[shamil-gadelshin]

Nice change! I added a small optional request for automation.

[shamil-gadelshin]

Is there a way to automate 8-9? We'll definitely miss these steps without some warning from CI or other tools.