fix: update Large hardware tier description to reflect current network scale

[cargopete]

Summary

The Large setup description states it is "Prepared to index all currently used Subgraphs" — this is no longer accurate and is misleading to indexers planning their infrastructure.

The problem:

• These figures were written when the network had far fewer subgraphs
• The network now has 15,000+ Subgraphs (as of 2026)
• Community operators report ~600 GB total RAM for ~1,700 Subgraphs (data from Marc-André / Ellipfra, April 2026)
• The Large tier (468 GB Postgres + 184 GB VMs = ~652 GB total) covers roughly 1,500–2,000 Subgraphs, not the full network

The fix:
Updated the Large tier description to accurately reflect ~1,500–2,000 Subgraph capacity and note that full-network indexing requires substantially more resources.

Supporting data

• Community operator data: ~600 GB RAM for ~1,700 Subgraphs (Marc-André | Ellipfra, April 2026)
• Analysis: https://www.lodestar-dashboard.com/blog/indexer-memory-1000-subgraphs

The table values (CPUs, disk) are left unchanged — only the description was inaccurate.