better reuse of lookups for inline contextual ligature substitutions

Currently we will create a new lookup for each inline contextual ligature substitution; that is, given the fea,

```fea
feature test {
sub f' i' by f_i;
sub f' f' by f_f;
} test;
```

we will generate three lookups: one contextual lookup and two ligature substitution lookups.

I'm not sure how commonly used this syntax is, and our behaviour _does_ match fonttools, but it doesn't match afdko (which combines the lookups).

So low priority, but another situation where some optimization is possible.