Remove aggregate graph store limit

[forhappy]

Summary

• remove the 2 GiB canonical-payload and 5 million record admission caps from current graph-index manifests
• retain independently bounded 256 KiB immutable objects, 16 MiB write batches, and bounded point/range reads
• validate large stores through streamed file digests and bounded tree traversal instead of whole-graph materialization
• preserve the legacy monolithic snapshot and portable JSON reader limits

Root cause

The graph-index already published content-addressed tree objects, but its manifest validator and canonical digest writer still applied legacy whole-payload limits. Store operations also reopened graph.json and exported the complete snapshot, reintroducing whole-document memory limits after publication.

Validation

• cargo fmt --all -- --check
• cargo clippy --workspace --lib --bins --locked -- -D warnings
• cargo test --workspace --lib --bins --locked
• cargo test -p compass-graph --test store_snapshot --locked
• cargo test -p compass-cli --test store_cli --locked
• scripts/check_product_boundary.sh
• scripts/qualify_code_graph_v1.sh --fixtures-only
• Orca default-low qualification: 866,633 nodes, 1,785,277 edges, 2,336,432,359-byte canonical graph, 95,763 store objects; publication, reopen validation, status, and bounded query all succeeded