perf(subgraph): reduce store operations to speed up indexing

## Goal

Increase subgraph indexing speed (backfill and steady-state) by cutting store operations per event.

## Sub-issues

**Phase 1 — remove dead work (small diffs, no product impact):**
- [ ] #135 Remove dead SumTree maintenance from piece handlers
- [ ] #136 Stop materializing ProvingWindow entities
- [ ] #137 Remove dead tx-input ABI parsing in `handlePiecesAdded`
- [ ] #138 Cut per-event store round-trips (`loadInBlock`, redundant existence check)

**Phase 2 — structural (schema changes, client updates):**
- [ ] #139 Migrate weekly/monthly activity metrics to native timeseries + `@aggregation`
- [ ] #140 Slim `EventLog` / `Transaction` entities and drop unused counters

Start with the six linked sub-issues (#135–#140), especially the piece handlers and handlePiecesAdded named in Phase 1. Review the indexing paths and each sub-issue's scope before changing anything. Done means the selected store operations are removed or reduced while backfill and steady-state indexing remain correct and faster.

Written by the indexing model from the issue text.