Lookup sync: don't penalize peers that extend children of long chains

[dapplion]

However, we then penalize peers who do the same action. This is inconsistent and can lead to penalties for honest peers. We should not penalize peers who extend known long chains. We could have to hashsets

• failed_chains: reject descendant lookups and penalize
• long_chains: reject descendant lookups and don't penalize

We should fix this in a future PR

Originally posted by @dapplion in https://github.com/sigp/lighthouse/pull/7118#discussion_r2128094198

[gitToki]

Hi @dapplion could I take this ?

[dapplion]

Yes @gitToki let me know if you need pointers

[jimmygchen]

It looks like this is happening on devnet-3, we fall behind and fallback to range sync, and penalise the peer the served us data.

[jimmygchen]

Closed by #8042