Fix NDCube.rebin for lookup-table extra coords

[nabobalis]

PR Description

NDCube.rebin had three problems with lookup-table extra coords:

• They were resampled with offset=0, so each new pixel got the value at the first pixel of its bin, while the rebinned WCS gives the bin centre. rebin now passes (bin_shape - 1) / 2 for lookup tables. WCS-backed extra coords keep 0, because ResampledLowLevelWCS treats the offset as a shift of the pixel edge. Nothing public tells the two apart, so this checks the private _lookup_tables.

• ExtraCoords.resample put the new grids in an object array. When they all have the same length (every 1-D cube, for example), numpy makes a 2-D object array and np.interp raises TypeError. They are now a list.

• TimeTableCoordinate.interpolate interpolated absolute MJD floats (about 0.6 µs resolution for current dates) and dropped reference_time. It now interpolates seconds from the first time and keeps reference_time.

import numpy as np
import astropy.units as u
from astropy.time import Time
from astropy.wcs import WCS

from ndcube import NDCube

wcs = WCS(naxis=2)
wcs.wcs.ctype = ["WAVE", "FREQ"]
wcs.wcs.cunit = ["m", "Hz"]

# Time table on array axis 1 at 60 s cadence, rebinned by 2 along that axis.
cube = NDCube(np.zeros((4, 6)), wcs=wcs)
cube.extra_coords.add("time", 1, Time("2026-01-01T00:00:00") + np.arange(6) * 60 * u.s)
r = cube.rebin((1, 2))
print(r.axis_world_coords(wcs=r.extra_coords)[0].isot)

# 1-D cube.
wave = WCS(naxis=1)
wave.wcs.ctype = ["WAVE"]
wave.wcs.cunit = ["m"]
cube1 = NDCube(np.zeros(6), wcs=wave)
cube1.extra_coords.add("wave2", 0, np.arange(6) * u.m)
r1 = cube1.rebin((2,))
print(r1.axis_world_coords(wcs=r1.extra_coords)[0])

main:

['2026-01-01T00:00:00.000' '2026-01-01T00:02:00.000'
 '2026-01-01T00:04:00.000']
Traceback (most recent call last):
  ...
TypeError: Cannot cast array data from dtype('O') to dtype('float64') according to the rule 'safe'

this branch:

['2026-01-01T00:00:30.000' '2026-01-01T00:02:30.000'
 '2026-01-01T00:04:30.000']
[0.5 2.5 4.5] m

AI Assistance Disclosure

AI tools were used for:

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding
• No AI tools were used

Regardless of AI use, the human contributor remains fully responsible for correctness, design choices, licensing compatibility, and long-term maintainability.

K.