fix: Resolve ValueError in distance matrix pivot & correct speed grid filtering

[pdml422]

This PR addresses two bugs encountered when evaluating a Multi-modal GTFS dataset (Bus + Metro/Subway with constant speeds) in Hanoi.
Bug 1 : Constant-speed routes being skipped in Headway calculation

• Issue: In the speed loop, the condition if gtfs_length_i == gtfs_length: continue acts as a premature optimization. If a route has a perfectly constant speed (e.g., a Metro line running at exactly 35km/h), it passes the speed_grid thresholds without reducing the gtfs_length. Consequently, the loop skips computation for these speeds, resulting in constant-speed modes being incorrectly assigned the lowest speed bin (e.g., 5km/h). Furthermore, overwriting gtfs_selection inside the loop breaks the filtering logic for subsequent iterations.
• Fix: Removed the gtfs_length_i == gtfs_length check to ensure computation runs for all valid thresholds. Introduced gtfs_selection_i to preserve the original DataFrame scope during filtering.

Bug 2 : ValueError: Index contains duplicate entries, cannot reshape during .pivot()

• Issue: When using a multi-modal dataset (e.g., mode_factor for Bus=0.8, Metro=1.0), rounding collisions occur. Two different stops from different modes can end up with the exact same stop_quality_grid value (e.g., 0.47) but slightly different final quality_grid values after distance multiplication. The previous drop_duplicates logic included 'quality_grid', which failed to drop these overlapping pairs, ultimately crashing the .pivot() function.
• Fix: Excluded 'quality_grid' from the subset in drop_duplicates(). Chained .sort_values('quality_grid') and used keep='last' to ensure that in the event of a rounding collision, the highest accessibility score is correctly retained for that grid intersection.

Testing: Tested successfully on a custom Hanoi GTFS feed containing both street-level buses and grade-separated Metro lines. The fix prevents the ValueError and accurately colors Metro lines based on their true speed.