Fix for gwcs inverse transform with a bounding_box

[nden]

This PR fixes a test where the input passed to the inverse transform is out of bounds. It should be merged after gwcs#498.
More specifically the footprint defined by the test WCS is [1, 49]*u.nm.
The input passed is in GHz and well outside this footprint. As a result the Tabular model in the WCS errors with "out of bounds" . This PR adjust the input values to be within the footprint.
As a side note, I found it is best to define Tabular model with bounds_error=False, fill_value=np.nan. In this case an error is not raised and the value returned when out of bounds is NaN which is more in line with the WCS philosophy.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.87%. Comparing base (f8a8d6a) to head (a67a083).
Report is 7 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1204      +/-   ##
==========================================
- Coverage   86.92%   86.87%   -0.05%     
==========================================
  Files          63       63              
  Lines        4572     4580       +8     
==========================================
+ Hits         3974     3979       +5     
- Misses        598      601       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rosteen]

Thanks - you're planning to update the pin from your branch to 0.22 once that's released, right? Probably should keep this as draft until then. @nden do you have any idea when 0.22 might release? We could really use a release that fixes the FITS deprecation error being raised in CI by GWCS due to astropy 7.0.

[pllim]

Is this a PR on gwcs?

[rosteen]

This is what I was asking about above as well.

[pllim]

Oh I think it is spacetelescope/gwcs#498 that is already merged

[pllim]

Cleaned up version in #1207