perf: split Iter & IterByColRange types into Tx and MutTxId versions

[Centril]

We're wasting work for the Tx version doing unnecessary bookkeeping in case there is a TxState, but there will never be one.
Meanwhile, the MutTxId version is also doing some minor bookkeeping for the case there won't be a TxState, but there always will be one.

In the case of Iter for TxId, this can just be TableScanIter, which is about 4x faster according to flamegraphs.