feat: structured selection audit log

Motivation

Selection decisions are logged across ~15 separate lines, correlated only by deployment_id and lost on log rotation. No single record captures why an indexer was picked.

Proposal

Per-line logging exists (#103) but is scattered; the consolidated record is the open work. Struck items are already logged today.

One structured JSON record per /select-indexers call with:

• Request params: deployment_id, chain_id, num_candidates, budget — existing/blocked still counts only, not identities
• Full candidate list: weighted score, sync status, pool — today only top-5 scores + pool counts
• Rejected + why (denylist, pending, declined, decentralisation) — today only aggregate counts
• Each pick: pool, indexer, score, decentralisation met / fallback
• Final group

Goal: reconstruct any selection from one record (stdout or shared volume), not scattered lines.

Start at the /select-indexers entry point and trace the existing selection logging, including the current top-five scores, pool counts, and aggregate rejection counts. Done means one structured JSON record per call contains the full candidate and rejection details needed to reconstruct the selection, written to stdout or the shared volume.