Clean up ML code

[JonasAlaif]

Left to do:

• Clean up code for synthetic terms, I'm thinking that all terms could be put in the same TiVec with synthetic terms wrapped in a transparent Synth<T>(T) which provides e.g. a function fn child_ids(&Synth<Term>) -> &[Synth<TermIdx>] and the parser can index fn index(&self, Synth<TermIdx>) -> &Synth<Term> (both using transmutes) and lastly there's a fn non_synth(&self, Synth<TermIdx>) -> Option<TermIdx> which unseals a Synth if the idx is lower than the last parsed idx.
• Clean up code for ML explanation/generalisation
  • The code still needs some cleanup but that can
    be done in a separate PR. One issue where we fail to generalise is in sequences-18.log. This is due to a sequence as so:

Seq_append(SIN, [⭐ → Seq_take(⭐, SUB)])
Seq_append(SIN, Seq_take([⭐ → Seq_take(⭐, SUB)], Seq_sub([⭐ → Seq_add(Seq_sub(⭐, LEN), LEN)], LEN)))
Seq_append(SIN, Seq_take(Seq_take([⭐ → Seq_take(⭐, SUB)], Seq_sub([⭐ → Seq_add(Seq_sub(⭐, LEN), LEN)], LEN)), Seq_sub(Seq_add(SUB, LEN), LEN)))
// A simplified sequence would be
g(x, y)
g(g(x, y), z)
g(g(g(x, y), z), z)

• Add textual explanation of MLs