Fixed sparse parachains finality handling in on-demand parachains relay

[svyatonik]

That's the issue I've found when working on #1418:

1. RialtoParachain -> Millau messages relay is seeing new message at RialtoParachain block RialtoParachainBlock=100. It asks on-demand parachains relay to relay RialtoParachainBlock=100;
2. on-demand relay relays RialtoBlock=150 first, which actually has head of RialtoParachainBlock=110. So it is some head from the future;
3. on-demand relay never returns any RialtoParachain blocks with number > 100 => RialtoParachainBlock=110 is never relayed and relay stucks.

The solution is to always explicitly read finalized parachain head from relayed relay chain block when switching from RelayingRelayHeader to RelayingParaHeader state. A good side effect of that is that now we don't need this headers_map_cache, originally introduced in the #1405

[svyatonik]

Merging unreviewed, noted in #1407

[serban300]

Maybe NotRequested or NotNeeded would be more accurate.

[svyatonik]

I've been looking at it as "The source client refuses to report parachain head hash at this moment." primarily. But yeah - your option also seems sensible to me :)

[serban300]

Rather related to #1484 . I would make this structure more generic, for example:

pub enum ExtOption<T> { // needs better name
	None,
	Some(T),
	Unavailable,
}

And in parachain_head() instead of having

    let mut para_hash_at_source = ParaHashAtSource::None;
    let mut para_header_number_at_source = None;
    ...

I would use ExtOption<HeaderId> in order to manage both variables at once:

let para_header_id = ExtOption::<HeaderId>::None;
...

[svyatonik]

Yeah, seems like a good idea. But we're using some non-trivial combintations there - e.g.

							para_hash_at_source = ParaHashAtSource::Unavailable;
							para_header_number_at_source = None;

So it shall be done with care (if done at all). Can you, please, file an issue - I could try to handle it a bit later? Thanks!

[serban300]

I was already experimenting with this. If it works out and if that's ok for you, I will publish a PR later.

[serban300]

The comments were addressed as part of #1531 . Marking the PR as reviewed.