solves #726: bug in default inverse mapping

[0Navin0]

# =============================================
# Following the same example described in the issue #726  
# after bug fix
# =============================================

In [23]: wa
Out[23]:
<SkyCoord (ICRS): (ra, dec) in deg
    (5.46165002, -72.12183946)>


In [29]: wcsobj.backward_transform
---------------------------------------------------------------------------
NotImplementedError                       Traceback (most recent call last)
File ~/Softwares/gwcs/gwcs/wcs/_pipeline.py:639, in Pipeline.backward_transform(self)
    638 try:
--> 639     backward = self.forward_transform.inverse
    640 except NotImplementedError as err:

File ~/anaconda3/envs/rdm_env_latest/lib/python3.14/site-packages/astropy/modeling/core.py:1383, in Model.inverse(self)
   1382 elif self._inverse is not None:
-> 1383     result = self._inverse()
   1384     if result is not NotImplemented:
.
.
.
.
NotImplementedError: No analytical or user-supplied inverse transform has been implemented for this model.

# Even when backward transform is not implemented in the original definition of the `wcsobj`, it should default successfully to `numerical_inverse` as shown below:
In [24]: wcsobj.world_to_pixel(wa)
Out[24]: (np.float64(3.999999999638021), np.float64(3.999999999540023))

In [25]: wcsobj.numerical_inverse(wa.ra.deg, wa.dec.deg)
Out[25]: (3.999999999638021, 3.999999999540023)

This PR addresses a bug that triggered when a backward_transform is not defined and default numerical_inverse needs to be used.

Tasks

• no-changelog-entry-needed

News fragment change types:

• changes/726.bugfix.rst: fixes an issue

[mcara]

I suppose tests added in #600 are missing the important block of code that was added in that PR, as well as romancal regression tests do not test for this usage.

Could you, please, update or add unit tests to cover this code?

[mcara]

Also, you are using main branch.

[0Navin0]

Also, you are using main branch.

Oh sorry. I am still learning my way around Github, so this happened by mistake. Should I create a new feature branch, and then again create a pull request? Any suggestion is appreciated.