MCS graph construction utils

[scal444]

We use a simpler graph representation than substruct, since we don't need repeated label/recurse cycles and the scope of matching is typically more limited.

PR 3 towards #221

[greptile-apps]

Greptile Summary

This PR adds the CSR graph construction utility (buildGraphFromEdges) and GPU-side packed bitmap match tables (MatchTableHost/MatchTableDevice/uploadPairMatchTables) that form the foundational data structures for the FMCS CUDA pipeline.

• mcs_graph.cpp: Builds a sorted, symmetric CSR adjacency list from an undirected edge list with validation for out-of-range vertices and self-loops. Two concerns (silent size_t→int truncation in numVertices/numEdges and no duplicate-edge detection) were raised in the prior review cycle.
• fmcs_match_tables.cuh/.cu: Defines host and device bitset tables for atom/bond compatibility and packs all pairs into a single contiguous AsyncDeviceVector<uint32_t> in one upload, with device pointers pre-computed from sequential packing offsets.
• Tests: Covers sorted symmetry, empty/disconnected graphs, error paths, word-boundary bit packing, and contiguous buffer layout — good breadth for a foundations layer.

Confidence Score: 5/5

Safe to merge — the new graph-building and match-table upload code is well-scoped and the test suite validates the key invariants.

The graph construction and bitset-table upload are straightforward additions with no shared mutable state or complex ownership. The two open concerns from the prior round (int truncation on vertex/edge counts and no duplicate-edge guard in buildGraphFromEdges) are known and tracked — neither affects correctness for real molecule sizes — and no new defects are introduced here.

Files Needing Attention: No files require special attention beyond the previously noted items in src/mcs/mcs_graph.cpp.

Important Files Changed

Filename	Overview
src/mcs/mcs_graph.cpp	New CSR graph builder with validation; duplicate-edge and int-overflow concerns flagged in prior review threads
src/mcs/fmcs_cuda/fmcs_match_tables.cuh	Defines host/device bitmap match-table types and upload interface; clean struct layout, well-typed bit arithmetic
src/mcs/fmcs_cuda/fmcs_match_tables.cu	Implements contiguous device-side packing of match tables; offset accounting and stream usage are correct
tests/test_fmcs_foundations.cu	Good coverage of graph construction and table upload; tests sort order, symmetry, error paths, word-boundary packing, and empty cases
src/mcs/CMakeLists.txt	New CMake target wiring mcs_graph.cpp and fmcs_match_tables.cu into mcs_fmcs_foundations; include dirs and link deps look correct

_{Reviews (4): Last reviewed commit: "Use project-rooted includes in fMCS foun..." | Re-trigger Greptile}

[greptile-apps]

numVertices and edges.size() are size_t, but the Graph struct stores int numVertices and int numEdges. The silent static_cast<int> truncates without any guard for values above INT_MAX. In practice molecule graphs are tiny, but an explicit check would make the expectation clear and catch misuse early.

Suggested change

	graph.numVertices = static_cast<int>(numVertices);
	graph.numEdges = static_cast<int>(edges.size());
	if (numVertices > static_cast<size_t>(std::numeric_limits<int>::max()) ||
	edges.size() > static_cast<size_t>(std::numeric_limits<int>::max())) {
	throw std::runtime_error("MCS graph vertex/edge count exceeds int capacity");
	}
	graph.numVertices = static_cast<int>(numVertices);
	graph.numEdges = static_cast<int>(edges.size());

[evasnow1992]

Changes look good to me. Thanks!

[scal444]

As discussed I'll wait until CI is green and rebase before merging.