BUG: avoid scalar statistics for structured Monte Carlo results

[ting-hong-shieh]

Describe the bug

MonteCarlo.set_processed_results() passes every collected result to NumPy's scalar aggregation functions. This gives structured outputs two inconsistent behaviors:

• equal-shaped lists and NumPy arrays are flattened and reported as one scalar distribution;
• ragged lists and empty result sets raise ValueError or IndexError.

The raw values are already preserved correctly in MonteCarlo.results. The failure occurs while constructing processed_results.

To reproduce

from rocketpy.simulation import MonteCarlo

monte_carlo = object.__new__(MonteCarlo)
monte_carlo.results = {"samples": [[1, 2], [3, 4]]}
monte_carlo.set_processed_results()

print(monte_carlo.processed_results["samples"])

At cb6106a717207dd8fc2dfe1446d80ff75022f21b, the result is:

(2.5, 2.5, 1.118033988749895, 1.075, 3.925)

A ragged value such as [[1], [2, 3]] instead raises:

ValueError: setting an array element with a sequence

An empty list reaches np.quantile and raises IndexError.

Expected behavior

Summary statistics should only be calculated when every observation is a real-valued scalar. Strings, lists, dictionaries, arrays, mixed values, booleans, and empty results should remain available in results, while the existing five-element processed_results entry should be:

(None, None, None, None, None)

This keeps the public tuple shape unchanged.

Proposed fix

Classify each result series before calling NumPy. Aggregate real scalars and skip scalar statistics for structured or non-numeric observations. Add regression cases for scalar NumPy values, strings, equal and ragged lists, dictionaries, arrays, mixed values, booleans, and empty results.

Environment

• Base SHA: cb6106a717207dd8fc2dfe1446d80ff75022f21b
• RocketPy 1.13.0
• Python 3.12.6
• NumPy 2.5.2
• SciPy 1.18.0
• pytest 9.1.1
• macOS 26.5.2, arm64