Unify metric names

[svyatonik]

closes #1056

History of this issue:

1. initially we have had separate relay processes for everything and every process has been using 'prefix' feature of prometheus::Registry. So e.g. headers relay from Rialto to Millau had prefix Rialto_to_Millau_Sync and metric names like best_block_numbers, resulting in final names like Rialto_to_Millau_Sync_best_block_numbers feed to Prometheus;
2. then we've introduced complex relays which have several relays running in single process => we can't use Registry prefix anymore. Because different relays have different prefixes - e.g. single process (means single Registry) may have relays with prefix Rialto_to_Millau_Sync and Millau_to_Rialto_Sync. But at the same ideally we should have had metrics with same final names exposed by the process - to avoid using different dashboard for different configurations. That's why prefixes were created directly with final names, without using Registry prefix;
3. additionally, we have standalone metrics, which are supposed to be only registered once per process. E.g. Polkadotstorage overhead metric is independent of relay that exposes it.

Right now the situation is as follows:

1. we are still using Registry prefix for standalone relays and using final metric names + no-registry-prefix for complex relays;
2. standalone metrics are sometimes exposed several times - e.g. CPU usage may be exposed without prefix (process_cpu_usage_percentage) and with prefix (Rialto_to_Millau_Sync_process_cpu_usage_percentage).

After this PR, we'll be using final metric names everywhere. We won't be using Registry prefix. The only metrics-related thing that'll need fix after this PR is that every relay will keep collecting data for its own copy of standalone metric values, even though only one metric will be exposed. This should be fixed in follow-up PRs.

Things to do here:

• verify dashboards in Rialto <> Millau deployments;
• verify Rococo <> Wococo messages relay (since it doesn't use conversion rates, which coming from metrics, which have been altered here) checked with altruistic relayer on Rialto <> Millau, because Rococo <> Wococo is too outdated.

[svyatonik]

// takes a bit longer than expected - initially I was going to replace Prometheus registry with our own struct that will essentially be a map of id -> standalone metric. But that was not a best idea, so now I'm moving all standalone metrics to the binary (sustrate-relay) level. Loops will only add their specific metrics, which do not need deduplication + values sharing.

[acatangiu]

LGTM