\ref{mfilter} handles complex filtering of matrices.
If the requested column~\tt{@col@} is marked as sequential with~\tt{@seq@},
a~$O(lg n)$ bisect algorithm will be used;
otherwise,
it will undergo a~$O(n)$ linear scan.
Perform an~$O(lg n)$ bisect on a data set.
This is intended to limit recursion on very large data sets (and
consequently will increase performance).
This will bisect up until a certain point (the gap),
unless it finds the value in question.
After finding the value,
it will perform an~$O(n)$ linear backward search to find the first
occurrence of the value.
If the value is not found,
it will halt at the gap and return the first index of the gap,
which we will consider its "best guess",
at which point a linear search can be performed by the caller to
determine if the value does in fact exist at all.
(The reason this operates so oddly is because of its caller;
we could rid the gap and make this a general-purpose function if need be.
Technically,
the gap is useless and saves $lg g$ steps,
which may be very small.)