Fix ConstrainedReschedule Rust port giving wrong node_start_time (#16186)

[bnayahu]

Fixes: #16186.

Three bugs in the Rust port of ConstrainedReschedule (PR #14883):

1. Delay duration ignored when target is provided: push_node_back computed new_t1q using target.get_duration("delay", …), which returns None → 0 because Delay is not in the target gate table. The Delay end-time was therefore this_t0 + 0 instead of this_t0 + delay_duration, causing a u64 underflow in the successor overlap calculation and silently setting the measure start time to a wrapped garbage value (160 dt instead of 272 dt). Fixed by always reading Delay duration from the instruction parameter first, falling back to the target only for non-Delay ops.

2. pulse_align/acquire_align swapped in the push_node_back call: run_constrained_reschedule passed the two alignment arguments in the wrong order. As a result acquire_align inside push_node_back received the pulse alignment value (1) and Measure/Reset were never shifted to the correct acquire-alignment boundary. Fixed by correcting the argument order.

3. u64 underflow in overlap arithmetic: new_t1q - next_t0q and t1c - t0c used plain subtraction; when the current node ends before the successor starts (no real overlap) this wraps to a large value and incorrectly shifts successors. Changed to saturating_sub.

Also adds is_measure_or_reset() helper to consolidate the three Measure/Reset checks and to handle Python-level Measure/Reset operations (stored as OperationRef::Instruction) that do not match the StandardInstruction::Measure/Reset pattern.

Adds regression tests in test_scheduling_padding_pass.py.

AI/LLM disclosure

• I didn't use LLM tooling, or only used it privately.
• I used the following tool to help write this PR description: Clause Code, with the draft Qiskit skills
• I used the following tool to generate or modify code: Clause Code, with the draft Qiskit skills

[CLAassistant]

All committers have signed the CLA.