refactor(fork-choice): unify ancestor-at-slot walks into one helper

[tcoratger]

What

The lstar fork choice hand-rolled the same "climb parent links up to the block at a target slot" walk in two places with slightly different stop conditions. This factors them into one private helper that returns the ancestor root at that slot, or None when no block sits there or the chain leaves the known tree.

• Attestation ancestry check: now resolves the block at the ancestor's slot and compares it to the ancestor's exact root.
• Head-update finalized walk: now resolves the finalized-slot root and rebuilds the checkpoint, falling back to the trusted anchor when the helper returns None.

The third parent-climbing loop (weight accumulation) is a different operation — it credits every block above a start slot rather than finding one ancestor — so it is intentionally left untouched.

Why

The divergent stop conditions across duplicated walks were a correctness hazard. Consolidating to one helper with a single documented stop condition removes that drift risk.

Equivalence argument (behavior-preserving)

The two consolidated walks are the same underlying lookup: "climb from a root until the block whose slot equals the target; return that root, else None." Edge cases verified to map identically:

• Ancestor slot above descendant slot: short-circuits to False (guard kept).
• Lands exactly on target slot: returns that root; matches the original root-equality test and checkpoint build.
• Slot skipped (no block at target on the chain): None -> ancestry False / head-update keeps the trusted anchor, same as the original's too-low landing.
• Chain leaves the known tree mid-climb: None -> same fallbacks; the original broke and took its else branch because the landed slot was still above the target.
• Head at or below the finalized slot: same anchor / checkpoint outcomes.

None is handled explicitly before any Bytes32 comparison, since the strict Bytes32 equality rejects comparison with None.

Verification

• just check: All checks passed.
• uv run fill --fork=lstar --clean -k fork_choice -n auto: all 118 fork-choice vectors regenerate and pass; the broader lstar vector set is byte-identical across hash seeds (determinism check passed).

🤖 Generated with Claude Code