Optimise TermList::skip_to()

[Olly Betts]

It occurred to me that for at least glass and the remote backend, the termlist classes are decoding prefix-compressed data and skip_to() could do less work than currently by taking advantage of knowing that as we advance to the skip point we don't actually need to update the current term beyond the point where it matches the target.

(For glass we would still need to track what the current length would be as we squeeze the wdf value into the reuse byte for some entries, but that's less work than updating the current term for every entry skipped over.)

Not 2.0.0 material at this point, but doesn't affect the ABI or remote protocol and not invasive so potential 2.0.x material.