perf(fork-choice): precompute per-chain attestation scores (FCR)

[twoeths]

Motivation

• Compute attestation scores in batch, following Lighthouse (sigp/lighthouse#8951).
• Also leverage state.getEffectiveBalanceIncrementsZeroInactive() to loop through validators fast (no per-index toValue() deserialization).

Description

• Replace per-block getAttestationScore with a per-chain precomputeChainAttestationScores + isOneConfirmedWithScore pair. One pass over validators walks each vote's parent chain until it lands on a canonical-chain block; a single suffix sum materialises the score for every block on the chain — O(V × depth + B) vs. the previous O(B × V × depth).
• Add unslashedActiveBalances to FastConfirmationBalanceSource (pre-zeroed for inactive + slashed) so the hot loop never touches the beacon state.
• Structured as 6 self-contained commits (see the commit log); the old isOneConfirmed / getAttestationScore / ensureVoteMaps path is removed in the final commit.

part of #8837

AI Assistance Disclosure

Designed and implemented with Claude Code (Opus 4.7).

[gemini-code-assist]

Code Review

This pull request optimizes the fast confirmation logic by introducing precomputeChainAttestationScores, which reduces the complexity of attestation score calculations from O(B * V * depth) to O(V * depth + B) through a single-pass validator walk and suffix sum. The review feedback suggests further performance refinements in this hot path, specifically replacing Map and standard Array with Int32Array and Float64Array to reduce hashing overhead and optimize memory usage.

[gemini-code-assist]

Using a Map<number, number> for indexToPosition in this hot path introduces significant hashing overhead, especially since it's queried up to $O(V \times \text{depth})$ times per tick (where $V$ is the validator count, e.g., 500k+). Since nodeIdx is an index into the protoNodes array, we can use a Int32Array initialized with -1 for much faster $O(1)$ access.

Suggested change

	const indexToPosition = new Map<number, number>();
	for (let pos = 0; pos < chain.length; pos++) {
	for (const nodeIdx of ctx.getNodeIndices(chain[pos])) {
	indexToPosition.set(nodeIdx, pos);
	}
	}
	const indexToPosition = new Int32Array(protoNodes.length).fill(-1);
	for (let pos = 0; pos < chain.length; pos++) {
	for (const nodeIdx of ctx.getNodeIndices(chain[pos])) {
	indexToPosition[nodeIdx] = pos;
	}
	}

[gemini-code-assist]

If indexToPosition is refactored to use an Int32Array, the lookup should be updated to check against the sentinel value.

Suggested change

	const hit = indexToPosition.get(cur);
	const hit = indexToPosition[cur];
	if (hit !== -1) {
	landedPos = hit;

[gemini-code-assist]

For performance-critical numeric arrays, using a TypedArray like Float64Array is generally preferred over a standard Array to ensure better memory layout and consistent performance in V8.

Suggested change

	const scoreAtPosition = new Array<number>(chain.length).fill(0);
	const scoreAtPosition = new Float64Array(chain.length);