[Frontend] vcix: lower fused matmul whose operand is vector_store'd, fix exp chunking

[YWHyuk]

Fixes SDPA, which was broken on this branch by the C++->Python vcix port (lower_to_vcix.py). Bisected decisively: C++ vcix passes SDPA, Python vcix does not; exp-chunking and dma-fine-grained were ruled out, leaving _lower_matmul.

1. Fused matmul left un-lowered (the numerical bug). _lower_matmul bailed on ATag is None or BTag is None -- i.e. it required an MVIN dma_start tag for both operands. In SDPA's scores·V matmul, operand B is the softmax output produced in place by affine.vector_store, not DMAed, so BTag stayed None and the matmul was left in the IR -> wrong attention output. The C++ MatmulOpLowering marks an operand initialized by either a dma_start or a preceding affine.vector_store into its root memref; the Python port omitted that branch. Restored it (isAInit/isBInit). BTag/BAsync stay None/0 and are only read under if BAsync:, so the B dma_wait is correctly skipped, matching C++.

2. n>1 transcendental chunking crash. _make_sf_vc_v_iv called vector.ExtractStridedSliceOp(offsets, sizes, strides, vec) -- wrong arg order, missing the result type and vector operand -> TypeError under these MLIR bindings. Fixed to (result=legal_ty, vector=vec, offsets, sizes, strides). Only large transcendentals (n>1, e.g. SDPA softmax exp) reach it, so CI's small-tile (n==1) tests never hit it.

Validated end-to-end (Spike+TOGSim allclose): SDPA 56 cases pass (was crash/wrong); matmul/bmm/conv2d regress clean. Pass-level: after the fix the Python vcix output is byte-identical to mlir-opt -test-pytorchsim-to-vcix for the SDPA kernel.

🤖 Generated with Claude Code

[YWHyuk]

Folded into #264 (cherry-picked, conflict resolved keeping both the C++-parity subtileK guard and the vector_store init branch). Closing this in favor of #264, which now carries the SDPA vcix fused-matmul + exp-chunk fix together with the other review fixes.