Beef up benchmarking machine

[ghzlatarev]

Currently our bare metal benchmark machine c5d.metal is actually less performant than the substrate reference hardware that we recommend:

+----------+----------------+--------------+--------------+-------------------+
| Category | Function       | Score        | Minimum      | Result            |
+=============================================================================+
| CPU      | BLAKE2-256     | 894.04 MiB/s | 1.00 GiB/s   | ❌ Fail ( 86.9 %) |
|----------+----------------+--------------+--------------+-------------------|
| CPU      | SR25519-Verify | 577.11 KiB/s | 666.00 KiB/s | ❌ Fail ( 86.7 %) |
|----------+----------------+--------------+--------------+-------------------|
| Memory   | Copy           | 4.81 GiB/s   | 14.32 GiB/s  | ❌ Fail ( 33.6 %) |
|----------+----------------+--------------+--------------+-------------------|
| Disk     | Seq Write      | 142.29 MiB/s | 450.00 MiB/s | ❌ Fail ( 31.6 %) |
|----------+----------------+--------------+--------------+-------------------|
| Disk     | Rnd Write      | 64.05 MiB/s  | 200.00 MiB/s | ❌ Fail ( 32.0 %) |
+----------+----------------+--------------+--------------+-------------------+

Rebenchmarking with a beefier machine that fits more closely to the reference hardware, will drop our runtime weights and improve performance. Low hanging fruit.