tp: add IndexScan, which looks rows up through a dataframe's index

Equalities on columns with an index, like a `CREATE PERFETTO INDEX` on
a track id, are answered by the dataframe interpreter by searching the
index. On the executor they cost a full scan per run. Before the
dataframe module's queries can move to the executor, it needs the same
lookup, finding the same rows in the same order.

IndexScan narrows the index's entries by each equality on its leading
columns, through filter::SearchIndexEqual as the interpreter does, and
reads the rows in the index's order. Nothing is sorted or copied: each
batch is a window of the dataframe's storage selecting a run of
increasing rows, so rows far apart take a batch each. Any other
condition is left to a Filter after the scan.

The differential test now builds indexes and runs index plans. No
callers yet.