Fix panic in pending_runtime_api with Aura consensus env (#814)

[tgmichel]

Rel MOON-1909

[4meta5]

This changes the error when the header of the best block cannot be unwrapped from:

for all runtime APIs
"Runtime api access error: {:?}", e

to

Cannot get header for block {:?}, block

[tgmichel]

This changes the user error when the header cannot be unwrapped from:

for all runtime APIs
"Runtime api access error: {:?}", e

to

Cannot get header for block {:?}

Before we were just assuming the header could be unwrapped (wrong) and now handle that error specifically. Runtime api access error message still the same.

[4meta5]

Thanks for explaining! As far as I understand

Before we were just assuming the header could be unwrapped (wrong) and now handle that error specifically.

We handle header not being unwrapped by returning the client error Cannot get header for block {:?}

Runtime api access error message still the same.

Now I think the runtime API access error is not emitted if the header cannot be unwrapped

[crystalin]

This should go after the timeout line.
Also why do we need to create a block for this ?

[tgmichel]

This cherry-picked solution from upstream uses parent hash for building the pending runtime api and, in the case of tests like this one, parent hash of genesis is not a valid block to initialize. Thus the need of creating one.

Tbh I don't like it and I can propose a better way of doing it upstream, which will not require modifying the tests like that.

[4meta5]

So this PR applies polkadot-evm#842 to our frontier fork

I'm still not sure what this has to do with pallet-randomness::on_initialize being introduced in 1802...

[tgmichel]

So this PR applies polkadot-evm#842 to our frontier fork

Not really, or not only. This PR avoids the unwrap thing, but also builds the pending block over parent instead of best. And that's the main issue:

api.initialize_block(&best, &header) VS. api.initialize_block(&parent_hash, &header)

[4meta5]

interesting, when is best != parent_hash? is it only when a node is syncing?

[tgmichel]

when is best != parent_hash?

Always. best is the tip of the canonical chain, and in this case parent is best's parent.