Feature: Accept image list as input to get_frame_cov (bypass MP4 decode)

[EYH0602]

Motivation

In game-fuzz, we record gameplay as MP4 at a fixed FPS, then pass the file to
get_frame_cov() which decodes it back into frames for perceptual hashing. This is
wasteful in two ways:

1. Encode-then-decode roundtrip — we capture screen frames, encode them into MP4,
   then gamecov decodes them right back into frames.
2. No control over which frames are hashed — the caller has no way to pass
   semantically meaningful frames (e.g., captured at action boundaries or during key
   game state transitions). All frames at the recording FPS are processed.

Games are continuous simulations — unlike browser/desktop automation, the game world
evolves between player inputs. An ideal sampling strategy would capture frames both at
action boundaries and periodically during idle/wait periods. This requires the ability
to pass pre-captured images directly.

Proposed API

Add a function that accepts a sequence of images instead of an MP4 path:

from gamecov import get_frame_cov_from_images, FrameCoverage
from PIL import Image
from returns.result import Result

def get_frame_cov_from_images(
    images: list[Image.Image],  # or list[np.ndarray]
) -> Result[FrameCoverage, str]:
    """Compute frame coverage from a list of pre-captured images."""
    ...

This would complement the existing get_frame_cov(path: str) which remains useful when
an MP4 is already available.

Use Case

# In the fuzzer's action execution loop:
screenshots: list[Image] = []
for action in seed.actions:
    execute(action)
    screenshots.append(capture_screen(region))

# Compute frame coverage directly from screenshots
frame_cov = get_frame_cov_from_images(screenshots)

Benefits

• Eliminates MP4 encode/decode overhead for coverage computation
• Enables action-boundary and event-driven frame sampling strategies
• Maintains backward compatibility (existing MP4 API unchanged)