Describe the bug
Calling rolling_filter with window=1 returns an empty DataArray instead of the original data unchanged. A window of 1 is a valid no-op filter and should return the input data as-is. No error or warning is raised, so the data loss is silent.

To Reproduce

import xarray as xr
import numpy as np
from movement.filtering import rolling_filter

data = xr.DataArray(np.random.rand(10), dims=["time"], coords={"time": np.arange(10)})
result = rolling_filter(data, window=1)
print(result.sizes)  # Frozen({'time': 0})
print(result.values)  # []

Expected behaviour
rolling_filter(data, window=1) should return the original data unchanged. A window of 1 means no smoothing is applied.

Suggested fix
Add a guard before the trim step (or even better, at the very beginning of the function to avoid unnecessary processing):

if window == 1:
    return data

Root cause: When window=1, half_window = window // 2 = 0. The trim step calls isel(time=slice(0, -0)) which is equivalent to slice(0, 0) in Python — returning an empty array instead of the full data.

Log file
N/A

Screenshots
N/A

Computer used:

• OS: Linux (Ubuntu)
• Version: 22.04
• Hardware specs: N/A

Additional context
This can silently break downstream trajectory metric computations if a user passes window=1 as a smoke test or minimal smoothing step before calling functions like compute_turning_angle.