fix: resolve multiple open issues (#11, #19, #20, #7/#12/#16/#23)

[Thmen]

Summary

This PR addresses several open issues in the EGVSR repository with code fixes and documentation improvements.

Issues Fixed

Fix #11: Circular import in LPIPS module (profile.sh fails with AttributeError)

Problem: Running profile.sh (or any code importing LPIPS) fails with:

AttributeError: module 'metrics.LPIPS' has no attribute 'models'

This was caused by a circular import chain: models/__init__.py → dist_model → networks_basic → metrics.LPIPS.models (same __init__.py still loading).

Solution: Extract utility functions (normalize_tensor, tensor2im, l2, etc.) from models/__init__.py into a dedicated util.py module. Update networks_basic.py and dist_model.py to use relative imports (from . import util) instead of absolute imports that trigger circular dependencies. The PerceptualLoss class now uses lazy import for dist_model.

Fix #20: NaN metrics when running test.sh

Problem: All metrics (PSNR, LPIPS, tOF) show as NaN with warnings about "Mean of empty slice".

Root causes & solutions:

1. metric_calculator.py only searched for .png files in GT directories, but data could be in .jpg format (the dataset loader already supported both). Fixed to accept png|jpg.
2. Empty metric lists were averaged without protection, producing NaN. Added guards to skip empty lists and show informative warnings instead.
3. Bare except: clauses in main.py silently swallowed all errors during metric computation, making debugging impossible. Replaced with except Exception as e: and proper logging.

Fix #19: test.sh usage unclear

Problem: Users didn't know how to properly invoke test.sh and profile.sh.

Solution: Added usage instructions, argument descriptions, examples, and input validation with helpful error messages to both scripts.

Fix #7, #12, #16, #23: Testing dataset download links broken

Problem: The Baidu Netdisk download link for testing datasets (Vid4, ToS3, Gvt72) is no longer available.

Solution: Updated README to mark the link as defunct and provide alternative guidance for obtaining the datasets.

Changes

File	Change
codes/metrics/LPIPS/models/util.py	New file: extracted utility functions
codes/metrics/LPIPS/models/__init__.py	Simplified to import from util.py, lazy import for dist_model
codes/metrics/LPIPS/models/dist_model.py	Use relative import for util
codes/metrics/LPIPS/models/networks_basic.py	Use relative import for util
codes/metrics/metric_calculator.py	Support jpg files, protect against empty lists
codes/main.py	Replace bare except with proper error handling
test.sh	Add usage docs, validation, error messages
profile.sh	Add usage docs, validation, error messages
README.md	Note about broken download links