refactor(ops): Split matmul.py into modular files

[m96-chan]

Problem

src/pygpukit/ops/matmul.py is 1981 lines with mixed concerns.

Current State

src/pygpukit/ops/
└── matmul.py  (1981 lines - everything)

Proposed Structure

src/pygpukit/ops/matmul/
├── __init__.py          (exports)
├── generic.py           (matmul, batched_matmul, transpose)
├── fp8.py               (FP8 GEMM variants)
├── nvf4.py              (NVF4 GEMM)
├── gemv.py              (GEMV operations)
├── grouped.py           (grouped GEMM for MoE)
├── quantize.py          (quantization helpers)
└── availability.py      (all *_available functions)

Benefits

• Each file < 400 lines
• Easier to find specific operations
• Aligned with refactor(native): Split ops_bindings.cpp into modular files #131 (native binding split)
• Aligned with refactor(api): Standardize GEMM/GEMV Python binding naming convention #130 (naming convention)

Related

• refactor(api): Standardize GEMM/GEMV Python binding naming convention #130 (naming convention)
• refactor(native): Split ops_bindings.cpp into modular files #131 (ops_bindings split)

[m96-chan]

Completed in PR #168