Raise ValueError instead of a deprecation warning for unrecognized spine positions - #20782

Overall, this looks good.

One issue I think needs fixing before this goes in: the hard-coded fallback positions in WCSAxes.reset_wcs (core.py, the ["b", "l"] and ["bltr", "bltr"] defaults used when coord_meta has no default_*_position keys) are only valid for RectangularFrame. Those keys are documented as optional in generic_transforms.rst, so a generic transform with a non-rectangular frame now fails at construction:

import astropy.units as u
import matplotlib.pyplot as plt
from astropy.visualization.wcsaxes import WCSAxes
from astropy.visualization.wcsaxes.frame import EllipticalFrame
from astropy.visualization.wcsaxes.tests.test_transform_coord_meta import DistanceToLonLat

fig = plt.figure()
coord_meta = {"type": ("longitude", "latitude"), "wrap": (180, None),
              "unit": (u.deg, u.deg), "name": ("lon", "lat")}
ax = WCSAxes(fig, [0.1, 0.1, 0.8, 0.8], transform=DistanceToLonLat(R=6e3),
             coord_meta=coord_meta, frame_class=EllipticalFrame)
# ValueError: Unrecognized position(s): ['b'], should be one of c/h/v.

On main this emits three deprecation warnings that the user cannot avoid other than by passing all three default_*_position keys. The same happens with RectangularFrame1D (['l'], should be one of b/t) and with custom frames. Maybe the fallbacks could depend on the frame class, e.g., keep b/l/bltr for RectangularFrame subclasses and use frame_class.spine_names (or "#") otherwise, as wcsapi.py already does? A regression test for the non-rectangular case would be good too.

A couple of minor things outside the diff:

• The set_ticks_position, set_ticklabel_position, and set_axislabel_position docstrings still say positions should be b/t/l/r, which isn't true for non-rectangular frames (pre-existing, but this seems like a good time to fix it).
• CoordinateHelper.tick_params applies the size/width/color settings before calling set_ticks_position, so e.g. ax.coords[0].tick_params(length=10, left=True) on a RectangularFrame1D now updates the tick size and then raises. This is a user error so it may be fine, but validating first would avoid a partial update.