Early Pruning optimization for RAPTOR transfer relaxation

[andrii-rohovyi]

Summary

Implement the Early Pruning optimization that reduces RAPTOR query time by breaking out of transfer iteration early when remaining (sorted) transfers cannot improve the solution. Following the discussion with @t2gran during the OTP meeting on April 7, 2026, this PR is submitted for review.

Issue

Closes #7470

Unit tests

• All existing OTP unit tests pass with the change.
• No new unit tests were added — the optimization is a performance improvement that preserves identical routing results.
• Manual verification: benchmarked on 1,000 random queries in the Norway (Oslo region) graph across 30/60/90 min max transfer durations. Results are 100% identical to baseline (0 differences in paths found).
• Performance improvement: +8.5% at 30 min, +8.2% at 60 min, +10.4% at 90 min transfer duration.
• No regression on the standard OTP Norway SpeedTest (45 test cases, identical path counts).

Documentation

• Design and rationale documented in code comments, referencing the paper: Rohovyi, Abuaisha, Walsh — "Early Pruning for Public Transport Routing", accepted at WCTR 2026 (https://arxiv.org/abs/2603.12592).
• The optimization can be toggled via system property otp.raptor.earlyPruning (enabled by default, set to "false" to disable).
• No new user-facing configuration options were added.