Fix XMR keygen blame identities and timeout reporting
Problem
Different encodings of the same public key can split matching XMR keygen failure reports into separate result groups.
Missing messages do not prove which participant caused a failure. Blaming a missing participant can wrongly accuse an honest node.
Failed results with no blamed nodes are also broadcast even though receivers reject them, causing unnecessary traffic and warning logs.
Changes
- Use the exact member keys from the keygen request when building and comparing blame.
- Log missing participants, missing votes and conflicting views as diagnostics.
- At timeout, name only peers observed sending conflicting votes.
- Skip failure broadcasts with no blamed nodes.
- Add tests that preserve the existing block on on-chain XMR failure reports.
Scope
This is a Bifrost-only MR targeting S00(!5026 (merged)). It improves failure-result matching and diagnostics without requiring a separate THORNode change.
It does not change THORNode consensus rules or penalty handling. Keysign and stored keyshares are unchanged. XMR failure forwarding remains disabled.
Policy and limits
Silence alone is not treated as proof of fault. This avoids blaming participants based on incomplete local information.
A participant that stays silent can still block keygen. This MR does not add a penalty for silence, guarantee removal of that participant, or solve selective withholding. Named blame remains local evidence, not proof that automatically causes an on-chain penalty.
Rollout
Deploy between ceremonies. All intended participants should run the same reviewed build before the next ceremony starts. Mixed-version ceremonies have not been validated.