OmniJet for Embed in ParticleTransformer

[jpata]

use OmniJet tokenizer + GPT backbone for the per-particle embedding in ParticleTransformer.

[Laurits7]

One version is addressed in #69. The two other implementations not done so far are the

• OmniJetBB encoding + ParT features (256+13[padded to 16])
• OmnijetBB encoding + ParT features embedding (256+256)

[Laurits7]

v2.1 and v2.2 addressed in PR #70

[Laurits7]

In PR #75 both the ParT embedding and self-attention layers were replaced by the OmniJet backbone embedding for a simpler model.