Fix rounding in demand placement

[networmix]

Summary

• avoid floating point drift when placing demand

Testing

• pytest --cov=ngraph --cov-fail-under=85 --cov-report term-missing

[Copilot]

Pull Request Overview

This PR enhances demand placement by introducing deterministic rounding to mitigate floating-point drift.

• Added a _round_float helper to snap tiny values to zero.
• Applied rounding to placed_demand, placed_now, and remaining in place().
• Imported math and MIN_FLOW to support the new logic.

Comments suppressed due to low confidence (2)

ngraph/lib/demand.py:110

• remaining is computed using the unrounded placed_now, which can reintroduce floating-point drift. Consider rounding placed_now first and then calculating remaining based on the rounded value.

remaining = to_place - placed_now

ngraph/lib/demand.py:27

• New rounding behavior in _round_float should have dedicated unit tests covering finite values, non-finite values, and the MIN_FLOW threshold to ensure correctness.

def _round_float(value: float) -> float:

[Copilot]

[nitpick] The precision value 12 is a magic number. Extract this into a named constant (e.g., ROUND_PRECISION) or make it configurable to improve clarity and maintainability.

Suggested change

	@staticmethod
	def _round_float(value: float) -> float:
	"""Round ``value`` to avoid tiny floating point drift."""
	if math.isfinite(value):
	rounded = round(value, 12)
	# Precision for rounding floating-point values
	ROUND_PRECISION: int = 12
	@staticmethod
	def _round_float(value: float) -> float:
	"""Round ``value`` to avoid tiny floating point drift."""
	if math.isfinite(value):
	rounded = round(value, Demand.ROUND_PRECISION)