Trace shifts updates

[segasai]

While investigating the persistent wavelength offsets in the psfs saved in EXTOFF,
I looked at the trace_shifts code and there are couple of small things that may need addressing. In the end the effect from those are tiny.

• One change is that the hw variable was used for both x and y when creating the LSF to convolve the external spectrum with. Previously hw was potentially very large (when the external spectrum had small wavelength sampling).
• Another change is using several fibers to compute the dy/dwavelength rather than using the central fiber ( the spread of slopes is ~0.5 %, so the effect of this is small)

[sbailey]

@julienguy please review this PR for inclusion in Matterhorn.

[Copilot]

Pull request overview

Updates wavelength-offset (external spectrum cross-correlation) handling in desispec.trace_shifts, aiming to reduce tiny systematic biases in the reference-spectrum convolution and make dy/dλ estimation less sensitive to a single fiber.

Changes:

• Add an explicit width parameter to _prepare_ref_spectrum() and use it to set the cross-dispersion (x-axis) integration window when building the LSF kernel.
• Compute dy/dλ using an average over multiple fibers instead of only a single “central” fiber.
• Minor logging formatting updates for readability.

Comments suppressed due to low confidence (1)

py/desispec/trace_shifts.py:672

• _prepare_ref_spectrum() now takes a new width parameter, but the docstring Arguments/Returns section doesn't mention what width represents or how it affects the convolution. Please update the docstring to document width (boxcar extraction width in pixels, and that hw_x=width//2 is used for cross-dispersion integration) so callers know how to keep extraction and reference LSF consistent.

def _prepare_ref_spectrum(ref_wave, ref_spectrum, psf, wave, mflux, nfibers, width=7):
    """
    Prepare the reference spectrum to be used for wavelength offset
    determination. Here we convolve it to the right LSF and rescale it
    to match the measured flux.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

[sbailey]

From the data telecon discussion, we agree that this update is small enough that we won't include it in Matterhorn, due to the risk of messing up some corner case. I'm going to move this to a draft PR so that we don't accidentally merge it.