air=True in grid_tools causing weird wavelength bug

Steps to reproduce:

import numpy as np
import Starfish
from Starfish.grid_tools import PHOENIXGridInterfaceNoAlpha

grid = PHOENIXGridInterfaceNoAlpha(wl_range=[900, 50000], air=False)
wl = grid.wl

print("Air is {}: Maximum change in wavelength {:.3f} AA".format(False, np.max(np.diff(wl))))
print("Air is {}: Minimum change in wavelength {:.3f} AA".format(False, np.min(np.diff(wl))))
print()

grid = PHOENIXGridInterfaceNoAlpha(wl_range=[900, 50000], air=True)
wl = grid.wl

print("Air is {}: Maximum change in wavelength {:.3f} AA".format(True, np.max(np.diff(wl))))
print("Air is {}: Minimum change in wavelength {:.3f} AA".format(True, np.min(np.diff(wl))))

Output

Air is False: Maximum change in wavelength 0.225 AA
Air is False: Minimum change in wavelength 0.006 AA

Air is True: Maximum change in wavelength 743.863 AA
Air is True: Minimum change in wavelength -1293.716 AA

Obviously, the wavelength grid should be monotonic. I'm not really sure what's causing this, but it seems to be related to applying the vacuum_to_air method to a larger range of wavelengths than just the optical.

Start by running the reproduction snippet for PHOENIXGridInterfaceNoAlpha with air=False and air=True, then inspect the grid_tools air conversion path, especially vacuum_to_air. Done means the air=True wavelength grid remains monotonic across the reported range and the existing reproduction no longer shows negative wavelength differences.