Add benchmarks for for import/validation/production

[skunert]

Closes #2048

This PR adds some benchmarks to work on. The requirements are specified in #2048.
The goal is to capture the current performance of block import/block production/block validation.

Transfers

First, we have a benchmark that fills a block with transfer_keep_alive transactions until the block is full.
This block is then built/imported/validated.

1313 transfer transactions per block.

Triecache Enabled:

Benchmark	Time (ms)	Throughput (Kelem/s)
Import	164.59	7.9772
Production w/o proof	192.13	6.8339
Production w/ proof	220.21	5.9624
Validation	-	-

Triecache Disabled:

Benchmark	Time (ms)	Throughput (Kelem/s)
Import	220.6	5.9638
Production w/o proof	261.81	5.0151
Production w/ proof	270.22	4.8580
Validation	263.20	4.9886

Compute/IO

In addition to the benchmarking of blocks with transfer extrinsics, I used the glutton pallet
to fully use the weight of a block with configurable compute/IO part. The glutton pallet will just "burn" the leftover weight
of the current block in the on_idle hook of the pallet. The weight consumed for PoV size and time
can be chosen using the glutton pallets extrinsics.

Note: Throughput is not available for this benchmark since we add no extrinsics.

Triecache Enabled:

Benchmark	compute = 100%, storage = 0% (ms)	compute = 100%, storage = 100% (ms)
Import	264.09	259.51
Production w/o proof	271.31	267.88
Production w/ proof	272.64	275.72
Validation	-	-

Triecache Disabled:

Benchmark	compute = 100%, storage = 0% (ms)	compute = 100%, storage = 100% (ms)
Import	270.0	274.73
Production w/o proof	267.13	268.32
Production w/ proof	264.57	276.60
Validation	271.59	284.26

---

Overall, these benchmark results raise some questions. For the benchmarks that use the glutton pallet, I was expecting the PoV size to have a much larger impact
on the execution time.

Edit: Switched out client block import with execute_block for a more fair comparison, results make more sense now. Block import is fastest, followed by block production. Validate block is slowest due to the constant decoding of storage values.

[melekes]

lgtm

[bkchr]

Generally looks good, I mainly left some nitpicks.

[bkchr]

Why don't you use this from the utils?

[skunert]

Yeah this is a bit of a compromise. I am using the test client for the validate block benchmarks and test service for block production and block import since I wanted to reuse the nice convenience features of the test client block builder.