Remove ignores that may apply to longer words containing a given word as a substring

[mrjacobbloom]

example: a -> una works, but its ignore rule also filters out as -> como

May require filtering out some lookbehind rules but not lookahead rules, and vise-versa. Be smart.

Will probably hecka slow down building features.fea, but should also lead to fewer rules per ligature and thus more words available

[mrjacobbloom]

Maybe features_inject could have a shape along the lines of

interface SubRule {
  length: number; // still required? Eh
  fromAsString: string; // easier to do a substring check than the equivalent for an array
  subRule: string;
  leftIgnore: string | null;
  rightIgnore: string | null;
}