Lower linalg.generic before MLIR-to-LLVM translation for math ops

[xlinsist]

Deliverables

• Submit a PR on preventing unlowered linalg.generic from reaching mlir-translate --mlir-to-llvmir in Triton RISC-V math op lowering.
• Include focused pytest evidence showing that python/examples/test_core.py::test_math_op[float32-log-x] no longer fails with a missing linalg dialect diagnostic.

Background

The current repository has a math lowering failure where final MLIR translation receives a linalg.generic op:

error: Dialect `linalg' not found for custom op 'linalg.generic'

Representative failure:

/tmp/.../ll.mlir:53:5: error: Dialect `linalg' not found for custom op 'linalg.generic'

This indicates the final translation boundary is receiving IR that has not been fully lowered to dialects accepted by the selected mlir-translate path.

Task Description

• Reproduce from the triton-riscv repository root:

cd python/examples
pytest -q 'test_core.py::test_math_op[float32-log-x]'

• Capture the final ll.mlir or equivalent IR and identify the pass sequence that leaves linalg.generic in the translation input.
• Fix the pass pipeline so this math op is lowered before mlir-translate, or use a translation context that intentionally supports the dialect if that is the correct architecture.
• Run the focused pytest case above and at least one adjacent math op such as exp, sqrt, or sin.
• In the PR description, include the before/after IR boundary, the pass ordering change if any, and the pytest commands run.

[xbwcan]

This issue was caused by linalgtovir and virtovector not supporting math.log.