Log UnknownBlock errors from runtime

[tdimitrov]

This PR is building upon paritytech/substrate#12707 and is related to paritytech/polkadot-sdk#793

The problem

All runtime calls first check the runtime version before issuing the actual runtime api call. If a call is made for a block which is pruned (e.g. because a better fork was finalized), the runtime api version can't be checked and NotSupported error is returned. This is misleading because the problem is different. The call IS supported but it can't be executed at a pruned block.

How this is solved

This PR adds explicit logging for UnknownBlock runtime errors so that it is clear why the runtime call is failing. Furthermore UnknownBlock variant is added to RuntimeApiError enum in node/subsystem-types/src/errors.rs. This allows any subsystem to handle such errors in its own way.

What this PR doesn't do

The PR doesn't fix the root cause (prevent executing logic on pruned blocks).

Sample logs

E.g. for {node=~"versi-validator-f.*"} |~ "UnknownBlock|discarded":

[paritytech-cicd-pr]

The CI pipeline was cancelled due to failure one of the required jobs.
Job name: test-linux-stable
Logs: https://gitlab.parity.io/parity/mirrors/polkadot/-/jobs/2315557

[ordian]

What is the rationale for using warns for such errors?

The PR doesn't fix the root cause (prevent executing logic on pruned blocks).

How is it possible to do that in general case? Wouldn't you have a TOCTOU problem?

[ordian]

I misunderstood the point of the issue was to make the error more explicit and not silence it. However, I am still not sure what would the point of warning the user about this if this error is expected to happen as we already observe in the logs.

[BradleyOlson64]

Looks clean!

[tdimitrov]

These logs doesn't contribute much to the understandably of the logs. I'm closing the PR.