Add numpy read/write for MgVideo/MgAudio and memory-based grid processing

[Copilot]

Adds the ability to read/write MgVideo and MgAudio objects as numpy arrays, initialize MgVideo from an array, and return grid images as arrays instead of writing to disk.

Changes

Bug fix: from_numpy path handling

When constructing MgVideo(filename=..., array=..., fps=..., path=...), the file was correctly written to path/filename but self.filename was left as the bare filename, causing an immediate FileNotFoundError in get_video(). Fixed by updating self.filename, self.of, and self.fex to the actual write path after FFmpeg completes.

Features (previously implemented, now tested and documented)

• MgVideo.numpy() – returns (array, fps) where array has shape (N, H, W, 3), BGR uint8
• MgAudio.numpy() – returns the audio waveform as a 1-D float array
• MgVideo(filename, array=array, fps=fps, path=path) – initializes from a numpy array, writing the encoded video to disk
• MgVideo.from_numpy(array, fps, target_name=None) – writes an array directly to a specified output path
• mg_grid(return_array=True) – returns the grid image as a numpy array without writing to disk

Usage

import musicalgestures

# Read frames as array
video = musicalgestures.MgVideo("input.mp4")
array, fps = video.numpy()           # shape: (N, H, W, 3), uint8 BGR

# Round-trip: write array back to a new MgVideo
new_video = musicalgestures.MgVideo(
    filename="output.avi",
    array=array,
    fps=fps,
    path="/some/output/dir",         # was broken before this fix
)

# Memory-based grid preview (no file written)
grid = video.grid(height=300, rows=3, cols=3, return_array=True)  # numpy array

Tests

Added tests/test_numpy.py with 18 cases covering all of the above, including the path-parameter regression.