Replace direct-DB validator deregistration with consensus-based removal

[raymondjacobson]

The deregisterValidator() function was directly deleting rows from core_validators outside of CometBFT consensus, causing state drift across nodes. Production nodes were reporting different chain state node counts because each node independently decided when to delete entries. Moreover depending on which node you state sync from, you may get different initial values in the validators table.

In this PR:

• Add a remove flag to the ValidatorDeregistration proto message to distinguish between jailing (underperformance) and removal (eth L1 deregistration, deletes row entirely)
• Introduces removeValidator() which submits a deregistration attestation with remove=true through CometBFT consensus, ensuring all nodes process the deletion deterministically
• Replaces both call sites (validator warden and eth event listener) that previously called deregisterValidator() with the new consensus-based removeValidator()
• Fixes self-unjailing: jailed nodes that come back online can now re-register through consensus to unjail themselves, which was previously blocked because isSelfAlreadyRegistered returned true for jailed nodes (this was a bug introduced in my previous PRs)
• Also fix the uptime page which would crash for all nodes if one node had no rollup info

[rickyrombo]

Was the problem that deregistered state was not going through consensus or jailing?

[raymondjacobson]

Was the problem that deregistered state was not going through consensus or jailing?

deregistered state was not going through consensus, at all! there was a function to just watch eth events coming in and then straight up remove the node