MCP project-info tools still page the whole symbol census (tracedecay_files, todos, port_status/order)

[ScriptedAlchemy]

#2140 removed whole-census loads from the dashboard graph reads: the interactive catalog now derives per-file symbol counts (CodeGraphCensusV1::largest_files), kinds, and languages when it is built. The MCP project-info handlers still page every symbol of the generation:

• crates/tracedecay-mcp/src/handlers/info/verified.rs::all_symbols calls graph.symbols_page(None, INFO_SYMBOL_CENSUS_LIMIT = 500_000) and refuses once a repo passes 500k symbols.
• indexed_files (the tracedecay_files tool) counts symbols per file from that census. largest_files already holds exactly this count: one per symbol whose binding has a logical_path, the same rule as required_file_path. VerifiedGraphQuery has no census() accessor yet.
• todos.rs (tracedecay_todos) and symbols_in_dir (port_status, port_order) filter the full census by path prefix and kind, where symbols_in_logical_files_page over the prefix's files would do.

On this repository (~200k symbols), every tracedecay_files call pages 200k symbols to count them.

Found by the legacy-api-sweep lane; outside its area (MCP project-info / graph-query).