Add reflected spectrum encoder

[wsnoble]

Try adding a second spectrum with (m_pre - m_i)/z_i as the m/z values, a la the Panda paper.

[bittremieux]

Should we provide a second spectrum as input, or just add the reflected peaks to the original input spectrum?

[wsnoble]

No, we should start by mimicking what is done in Pandanovo, which is to create a separate encoder for the reflected spectrum. It may be worth also trying the simpler approach subsequently.