Add visqol-python

[talker93]

Adds visqol-python, a port of Google's ViSQOL (objective audio/speech quality metric, MOS-LQO scale 1–5) that installs from PyPI without Bazel.

Why awesome:

• Useful across several research fields: audio codec eval, speech enhancement, VoIP/telephony, generative audio (e.g. AudioCraft METRICS already use ViSQOL).
• Bit-near parity with the C++ binary: audio conformance ≤ 0.0001 MOS diff (10/10 cases), speech polynomial 0.001, speech lattice 0.002.
• Optional [lattice] extra runs the same TFLite deep-lattice quality mapper as ./bazel-bin/visqol --use_lattice_model=true (the C++ default), via Google's ai-edge-litert.
• Optional Numba acceleration gets RTF down to ~0.064 (~9× over scipy).
• Apache-2.0; CI on Python 3.9–3.13 × NumPy 1.x/2.x.

Inserted alongside the existing pypesq / pystoi entries — these are the three commonly used objective speech-quality metrics.

Re-read the contributing guidelines:

• ✅ audio/music related (objective quality metric)
• ✅ open source license that allows contributions (Apache-2.0)
• ✅ useful for several fields (codec eval / VoIP / speech enhancement / generative audio)
• ✅ listed on PyPI (📦 link included)
• ✅ open development on GitHub
• ✅ Python 3 only