`Combiner.sigma_clipping`: `low_thresh`/`high_thresh` docstring no longer matches behaviour (`None` means 3σ, negatives are not `abs()`-ed)

[mwcraig]

The sigma_clipping docstring (ccdproc/combiner.py, currently lines 443–452) still says:

low_thresh : positive float or None, optional
Threshold for rejecting pixels that deviate below the baseline value.
If negative value, then will be convert to a positive value.
If None, no rejection will be done based on low_thresh. Default is 3.

high_thresh : positive float or None, optional
Threshold for rejecting pixels that deviate above the baseline value.
If None, no rejection will be done based on high_thresh. Default is 3.

That describes the original ccdproc implementation. In git show v2.0.1:ccdproc/combiner.py the method does

if low_thresh is not None:
    if low_thresh < 0:
        low_thresh = abs(low_thresh)
    mask = (self.data_arr - baseline < -low_thresh * dev)
    ...
if high_thresh is not None:
    mask = (self.data_arr - baseline > high_thresh * dev)
    ...

so None really did mean "skip that side", and a negative low_thresh (only low_thresh) was made positive.

Since 2.4.0 (5d2abc0, "Always use astropy's sigma clipping and default to string-based functions", #794; the use_astropy option itself arrived in 2.2.0 with d57ca13) the method forwards the thresholds straight to astropy.stats.sigma_clip(..., sigma_lower=low_thresh, sigma_upper=high_thresh, ...). astropy's SigmaClip.__init__ (astropy 8.0.1, astropy/stats/sigma_clipping.py lines 175–176) does

self.sigma_lower = sigma_lower or sigma
self.sigma_upper = sigma_upper or sigma

with sigma=3 by default, so today:

• low_thresh=None / high_thresh=None clip at 3σ on that side rather than not clipping;
• low_thresh=0 / high_thresh=0 likewise become 3σ instead of rejecting everything off-centre;
• a negative threshold is passed through unchanged, which turns the comparison inside out (nothing is ever rejected on that side) rather than being converted to a positive value.

None of this is covered by the test suite, and the docstring was not updated in #794.

Related: #1001 (array-API fallback for sigma_clipping) deliberately reproduces the astropy semantics on non-NumPy namespaces so all backends agree, and changes the docstring to say that None is treated as 3; the negative-threshold wording and the 0 case are left as-is there. #1000 is the companion sigma_func change. Both are part of #929.