Find the start position of an instance of any of the given term
types, or return null when none of them are found.
Note: this is only reliable when there is at least some
state that unambiguously matches the given rule on the stack.
I.e. if you have a grammar like this, where the difference
between a and b is only apparent at the third token:
a { b | c }
b { "x" "y" "x" }
c { "x" "y" "z" }
Then a parse state after "x" will not reliably tell you that
b is on the stack. You can pass [b, c] to reliably check
for either of those two rules (assuming that a isn't part of
some rule that includes other things starting with "x").
When before is given, this keeps scanning up the stack until
it finds a match that starts before that position.
Note that you have to be careful when using this in tokenizers,
since it's relatively easy to introduce data dependencies that
break incremental parsing by using this method.