feat(benchmarking): tune benchmarking for non-return syscalls

[tltsutltsu]

Resolves: #2938

There are a couple of benchmarks preceded with this comment:

// We cannot call `gr_wait` multiple times. Therefore our weight determination is not 
// as precise as with other APIs. 

Lets call them "one-time benchmarks"

This PR resolves this problem by splitting the benchmarking process of pallet_gear into two runs. The first run is made with default parameters (--steps=20, --repeat=50) for all extrinsics, excluding one-time ones. The second run is made only for one-time extrinsics, with --steps=2, --repeat=1000 params. This allows to evaluate one-time benchmarks more thoroughly.

Here's how it's done, technically:

• run_all_benchmarks.sh obtains a list of all extrinsics.
• run_all_benchmarks.sh runs all benchmarks, excluding one-time ones. The output is written into the pallet_gear.rs file.
• run_all_benchmarks.sh runs only one-time benchmarks. The output is written into the pallet_gear_onetime.rs file.
• merge-outputs.sh moves WeightInfo trait definitions and weights from the pallet_gear_onetime.rs file to main pallet_gear.rs file.

No changes needed in CI, and everything, including weights unpacking script, is are already working fine.
Successful benchmark action run / Weights diff table

The set of these one-time benchmarks can be changed anytime without need of modifying the logic in the script. Additionally , merge-outputs.sh also allows to merge more than one file by specifying multiple files.

@gear-tech/dev

[ukint-vs]

Could you post a table here with the weight changes?

[tltsutltsu]

@ukint-vs

Could you post a table here with the weight changes?

Sure!

name	master	tuned	diff
gr_reply_commit	20.8 µs	16.7 µs	+24.61%
gr_reply_commit_wgas	16.0 µs	23.4 µs	-31.69%
gr_reservation_reply	11.1 µs	10.2 µs	+8.71%
gr_reservation_reply_commit	5.3 µs	10.3 µs	-48.34%
gr_reply	16.8 µs	24.5 µs	-31.20%
gr_reply_wgas	20.5 µs	21.7 µs	-5.37%
gr_reply_input	21.9 µs	20.2 µs	+8.30%
gr_reply_input_wgas	0 ps	30.9 µs	N/A
gr_exit	23.7 µs	181.1 µs	-86.92%
gr_leave	12.2 µs	172.9 µs	-92.95%
gr_wait	13.2 µs	149.0 µs	-91.17%
gr_wait_for	14.0 µs	185.5 µs	-92.46%
gr_wait_up_to	12.2 µs	167.1 µs	-92.68%

These weights were measured with 1000 repeats.

Here is a full table with other weights