Make CellCollection.select() raise on invalid float at_most - #3860

Pre-PR Checklist

• This PR is a bug fix, not a new feature or enhancement.

Summary

CellCollection.select() still has the fraction check that #3853 just fixed in AbstractAgentSet.select(), so invalid float values for at_most are silently accepted on cell collections: nan selects every cell, 1.5 selects two cells, and -0.5 or 0.0 select none.

Bug / Issue

Same root cause as #3852, in mesa/discrete_space/cell_collection.py:

if at_most <= 1.0 and isinstance(at_most, float):
    at_most = int(len(self) * at_most)

nan <= 1.0 is False, so NaN skips the branch and reaches the generator, where count >= at_most is never true and the loop never stops. A float above 1.0 also skips the branch and ends up used as a raw count, so at_most=1.5 gives 2 cells rather than an error. Negative floats and 0.0 go through the branch and floor to a value that selects nothing.

On a collection of 10 cells at current main I get nan -> 10, 1.5 -> 2, -0.5 -> 0, 0.0 -> 0.

Implementation

I mirrored the check from #3853: on the float path (excluding the inf default), at_most must be in (0.0, 1.0], otherwise ValueError is raised. Integers and valid fractions are unchanged, and the docstring gets a Raises: entry.

Testing

I added the same four cases as #3853 (nan, 0.0, -0.5, 1.5) to the existing test_cell_collection. With only the test applied it fails with DID NOT RAISE ValueError; with the fix it passes. After the fix, 1.0 -> 10, 0.5 -> 5 and 3 -> 3 still behave as before. The whole tests/discrete_space directory passes (100 tests), and ruff check and ruff format --check are clean on both files.

Additional Notes

This makes CellCollection.select() and AgentSet.select() consistent again. I didn't find any caller inside the repo that passes a float above 1.0 to CellCollection.select().

Found by a defect-hunting pipeline I build and run (Dev-next-gen), using Claude Code with Anthropic's Claude Opus 5.