feat: add GraphPartitioner — balanced k-way graph partitioning

Three partitioning strategies:
- BFS-based: grows partitions from farthest-first seed vertices
  using round-robin breadth-first expansion for balanced sizes
- Kernighan-Lin: BFS initial + iterative vertex pair swaps to
  minimize edge cuts (extends to k-way via pairwise refinement)
- Spectral bisection: Fiedler vector (2nd Laplacian eigenvector)
  via power iteration + deflation, recursive bisection for k-way,
  refined with KL swaps

PartitionResult provides:
- Per-vertex partition assignment
- Edge cut count and cut ratio
- Imbalance ratio (max partition size / ideal size)
- Per-partition metrics: size, internal/external edges, conductance
- Members list per partition
- Formatted toString() summary

729 lines source + 465 lines tests (36 tests). All tests pass.

Author: sauravbhattacharya001