Background and Extract not respecting disp_axis correctly

When trying to use the non-default `disp_axis` and `crossdisp_axis`, @fchabourbarra and I are getting a variety of broadcasting errors, and results that are nonsense. Our workaround right now is transposing our data, which isn't ideal. The biggest issue is the case where it apparently extracts correctly (i.e. without complaint), but extracts along the wrong axis.

I've attached a (lengthy) minimum working example that tests _I think_ most of the failure modes

````python
import numpy as np
from astropy.nddata import CCDData
import astropy.units as u
from specreduce.tracing import ArrayTrace
from specreduce.background import Background
from specreduce.extract import BoxcarExtract

# Making some fake data that is "vertical" (swapping disp/cross axes from default)
data = np.zeros((300,100)) # disp_axis=0, crossdisp_axis=1
data[[120,239],:] = 1 # put 2 lines at fixed wavelengths
img = CCDData(data, unit=u.adu)

print('img:', img.shape)

trace_array = np.ones(300)*75 # a line at crossdisp_axis = 75, has length=300
tr_bad = ArrayTrace(img, trace_array)
print('bad trace:', len(tr_bad.trace)) # should = 300

# if we transpose the data, ArrayTrace does it right
tr = ArrayTrace(img.data.T, trace_array)
print('trace:', len(tr.trace)) # should = 300

# this raises the error:
# ValueError: could not broadcast input array from shape (100,) into shape (300,)
# bk = Background(img, tr, disp_axis=0, crossdisp_axis=1)

# this also raises the error:
# ValueError: could not broadcast input array from shape (100,) into shape (300,)
# bk = Background(img.data.T, tr, disp_axis=0, crossdisp_axis=1)

# setting axes back to default (which is WRONG for image, and doesn't match trace)
# this... works, but gives wrong length (100)
bk = Background(img, tr, disp_axis=1, crossdisp_axis=0)
print('bad bkgd_spectrum:',bk.bkg_spectrum().spectral_axis.shape)

# transposing the image fixes it
bk = Background(img.data.T, tr, disp_axis=1, crossdisp_axis=0)
print('bkgd_spectrum:',bk.bkg_spectrum().spectral_axis.shape)

# this appears to work
extract = BoxcarExtract(img, tr, disp_axis=0, crossdisp_axis=1)
# but this raises error:
# ValueError: could not broadcast input array from shape (100,) into shape (300,)
# print(extract.spectrum.flux.shape)

# this appears to work, with WRONG axes that don't match trace
extract = BoxcarExtract(img, tr, disp_axis=1, crossdisp_axis=0)
# and gives wrong length (100)
print('bad extract:', extract.spectrum.flux.shape)

# transposing works
extract = BoxcarExtract(img.data.T, tr, disp_axis=1, crossdisp_axis=0)
# and gives wrong length (100)
print('extract:', extract.spectrum.flux.shape)

````

which outputs this:

```
img: (300, 100)
bad trace: 100
trace: 300
bad bkgd_spectrum: (100,)
bkgd_spectrum: (300,)
bad extract: (100,)
extract: (300,)
```

Start by running the attached minimum working example, then inspect ArrayTrace, Background, and BoxcarExtract where disp_axis and crossdisp_axis are accepted. Compare behavior for both axis orientations and use the reported broadcasting errors and spectrum shapes as the checks; done means non-default axes produce the expected 300-element results without transposing the input.

Written by the indexing model from the issue text.