Rewrite Metric_computation with parallel processing and improved pipeline

[smishra3]

Summary

• Parallelize area-at-the-glass computation across movies using joblib with retry logic and exponential backoff for network resilience
• Fix SOX2 half-maximal calculation to use first timepoint intensity (biologically meaningful baseline) instead of max(), and correctly convert timepoints to hours
• Add fixed-cell (single timepoint) image support, metadata-only movie inclusion, and CLI arguments (--local, --local-csv) for local manifest loading
• Update io.py to support load_from_aws/local_path parameters; add joblib>=1.3.0 dependency; improve README Step 2 with gene metric definitions

Details
Parallel processing: add_bottom_mip_migration now processes movies in parallel via joblib.Parallel instead of sequential tqdm loop. Each movie is handled by _process_single_movie_area with up to 5 retries and exponential backoff (1s, 2s, 4s, 8s, 16s) for S3 network errors.

Bug fixes:

• SOX2 half-maximal: old code used max(df_id.int_smooth.values[0]) which calls max() on a scalar; new code uses
  df_id.int_smooth.values[0] (first timepoint value) as the upper bound of the dynamic range, and converts to hours with *(30/60)
• Fixed-cell images: detects single-timepoint images (dims.T == 1) and always indexes T=0 instead of the computed timepoint index

Pipeline improvements:

• Vectorized Z-plane normalization (replaces slow apply + lambda)
• Reduced memory in area computation by passing only needed columns to groupby
• Migration onset times rounded to nearest 0.5 hour for alignment with time grid
• Output columns reordered: key analysis columns first, then metadata
• Multi-channel intensity columns (Channel 2, Channel 3) included in output
• Movies without All Cells Mask added as metadata-only rows
• CSV saved without row index (index=False)

Test plan

• Verify CI passes (lock-check and requirements-check)
• Confirm Metric_computation.py imports succeed
• Verify --local CLI flag works for local manifest loading
• Compare output Image_analysis_extracted_features.csv against previous version to confirm expected differences (SOX2 values, hours conversion, additional columns)